[Doc] add common case for long waiting time (#5430)

Co-authored-by: Antoni Baum <antoni.baum@protonmail.com>

.buildkite/check-wheel-size.py

@@ -0,0 +1,36 @@
+import os
+import zipfile
+
+MAX_SIZE_MB = 200
+
+
+def print_top_10_largest_files(zip_file):
+    with zipfile.ZipFile(zip_file, 'r') as z:
+        file_sizes = [(f, z.getinfo(f).file_size) for f in z.namelist()]
+        file_sizes.sort(key=lambda x: x[1], reverse=True)
+        for f, size in file_sizes[:10]:
+            print(f"{f}: {size/(1024*1024)} MBs uncompressed.")
+
+
+def check_wheel_size(directory):
+    for root, _, files in os.walk(directory):
+        for f in files:
+            if f.endswith(".whl"):
+                wheel_path = os.path.join(root, f)
+                wheel_size = os.path.getsize(wheel_path)
+                wheel_size_mb = wheel_size / (1024 * 1024)
+                if wheel_size_mb > MAX_SIZE_MB:
+                    print(
+                        f"Wheel {wheel_path} is too large ({wheel_size_mb} MB) "
+                        f"compare to the allowed size ({MAX_SIZE_MB} MB).")
+                    print_top_10_largest_files(wheel_path)
+                    return 1
+                else:
+                    print(f"Wheel {wheel_path} is within the allowed size "
+                          f"({wheel_size_mb} MB).")
+    return 0
+
+
+if __name__ == "__main__":
+    import sys
+    sys.exit(check_wheel_size(sys.argv[1]))

.buildkite/nightly-benchmarks/kickoff-pipeline.sh

@@ -0,0 +1,26 @@
+#!/usr/bin/env bash
+
+set -euo pipefail
+
+# Install system packages
+apt update
+apt install -y curl jq
+
+# Install minijinja for templating
+curl -sSfL https://github.com/mitsuhiko/minijinja/releases/latest/download/minijinja-cli-installer.sh | sh
+source $HOME/.cargo/env
+
+# If BUILDKITE_PULL_REQUEST != "false", then we check the PR labels using curl and jq
+if [ "$BUILDKITE_PULL_REQUEST" != "false" ]; then
+  PR_LABELS=$(curl -s "https://api.github.com/repos/vllm-project/vllm/pulls/$BUILDKITE_PULL_REQUEST" | jq -r '.labels[].name')
+
+  if [[ $PR_LABELS == *"perf-benchmarks"* ]]; then
+    echo "This PR has the 'perf-benchmarks' label. Proceeding with the nightly benchmarks."
+  else
+    echo "This PR does not have the 'perf-benchmarks' label. Skipping the nightly benchmarks."
+    exit 0
+  fi
+fi
+
+# Upload sample.yaml
+buildkite-agent pipeline upload .buildkite/nightly-benchmarks/sample.yaml

.buildkite/nightly-benchmarks/sample.yaml

@@ -0,0 +1,39 @@
+steps:
+  # NOTE(simon): You can create separate blocks for different jobs
+  - label: "A100: NVIDIA SMI"
+    agents:
+      queue: A100
+    plugins:
+    - kubernetes:
+        podSpec:
+          containers:
+          # - image: us-central1-docker.pkg.dev/vllm-405802/vllm-ci-test-repo/vllm-test:$BUILDKITE_COMMIT
+          # TODO(simon): check latest main branch or use the PR image.
+          - image: us-central1-docker.pkg.dev/vllm-405802/vllm-ci-test-repo/vllm-test:45c35f0d58f4508bf43bd6af1d3d0d0ec0c915e6
+            command:
+            - bash -c 'nvidia-smi && nvidia-smi topo -m && pwd && ls'
+            resources:
+              limits:
+                nvidia.com/gpu: 8
+            volumeMounts:
+            - name: devshm
+              mountPath: /dev/shm
+          nodeSelector:
+            nvidia.com/gpu.product: NVIDIA-A100-SXM4-80GB
+          volumes:
+          - name: devshm
+            emptyDir:
+              medium: Memory
+  # TODO(simon): bring H100 online
+  # - label: "H100: NVIDIA SMI"
+  #   agents:
+  #     queue: H100
+  #   plugins:
+  #   - docker#v5.11.0:
+  #       image: us-central1-docker.pkg.dev/vllm-405802/vllm-ci-test-repo/vllm-test:45c35f0d58f4508bf43bd6af1d3d0d0ec0c915e6
+  #       command:
+  #       - bash -c 'nvidia-smi && nvidia-smi topo -m'
+  #       propagate-environment: true
+  #       ipc: host
+  #       gpus: all
+

.buildkite/run-amd-test.sh

@@ -1,38 +1,73 @@
-# This script build the ROCm docker image and run the API server inside the container.
+# This script runs test inside the corresponding ROCm docker container.
-# It serves a sanity check for compilation and basic model usage.
 set -ex
 
 # Print ROCm version
+echo "--- ROCm info"
 rocminfo
 
-# Try building the docker image
+# cleanup older docker images
-docker build -t rocm -f Dockerfile.rocm .
+cleanup_docker() {
+  # Get Docker's root directory
+  docker_root=$(docker info -f '{{.DockerRootDir}}')
+  if [ -z "$docker_root" ]; then
+    echo "Failed to determine Docker root directory."
+    exit 1
+  fi
+  echo "Docker root directory: $docker_root"
+  # Check disk usage of the filesystem where Docker's root directory is located
+  disk_usage=$(df "$docker_root" | tail -1 | awk '{print $5}' | sed 's/%//')
+  # Define the threshold
+  threshold=70
+  if [ "$disk_usage" -gt "$threshold" ]; then
+    echo "Disk usage is above $threshold%. Cleaning up Docker images and volumes..."
+    # Remove dangling images (those that are not tagged and not used by any container)
+    docker image prune -f
+    # Remove unused volumes
+    docker volume prune -f
+    echo "Docker images and volumes cleanup completed."
+  else
+    echo "Disk usage is below $threshold%. No cleanup needed."
+  fi
+}
 
-# Setup cleanup
+# Call the cleanup docker function
-remove_docker_container() { docker rm -f rocm || true; }
+cleanup_docker
-trap remove_docker_container EXIT
-remove_docker_container
 
-# Run the image
+echo "--- Resetting GPUs"
-docker run --device /dev/kfd --device /dev/dri --network host --name rocm rocm python3 -m vllm.entrypoints.api_server &
 
-# Wait for the server to start
+echo "reset" > /opt/amdgpu/etc/gpu_state
-wait_for_server_to_start() {
-    timeout=300
-    counter=0
 
-    while [ "$(curl -s -o /dev/null -w ''%{http_code}'' localhost:8000/health)" != "200" ]; do
+while true; do
-        sleep 1
+        sleep 3
-        counter=$((counter + 1))
+        if grep -q clean /opt/amdgpu/etc/gpu_state; then
-        if [ $counter -ge $timeout ]; then
+                echo "GPUs state is \"clean\""
-            echo "Timeout after $timeout seconds"
                 break
         fi
 done
-}
-wait_for_server_to_start
 
-# Test a simple prompt
+echo "--- Building container"
-curl -X POST -H "Content-Type: application/json" \
+sha=$(git rev-parse --short HEAD)
-    localhost:8000/generate \
+image_name=rocm_${sha}
-    -d '{"prompt": "San Francisco is a"}'
+container_name=rocm_${sha}_$(tr -dc A-Za-z0-9 < /dev/urandom | head -c 10; echo)
+docker build \
+        -t ${image_name} \
+        -f Dockerfile.rocm \
+        --progress plain \
+        .
+
+remove_docker_container() {
+   docker rm -f ${container_name} || docker image rm -f ${image_name} || true
+}
+trap remove_docker_container EXIT
+
+echo "--- Running container"
+
+docker run \
+        --device /dev/kfd --device /dev/dri \
+        --network host \
+        --rm \
+        -e HF_TOKEN \
+        --name ${container_name} \
+        ${image_name} \
+        /bin/bash -c "${@}"
+

.buildkite/run-benchmarks.sh

@@ -9,10 +9,10 @@ cd "$(dirname "${BASH_SOURCE[0]}")/.."
 (which wget && which curl) || (apt-get update && apt-get install -y wget curl)
 
 # run python-based benchmarks and upload the result to buildkite
-python3 benchmarks/benchmark_latency.py 2>&1 | tee benchmark_latency.txt
+python3 benchmarks/benchmark_latency.py --output-json latency_results.json 2>&1 | tee benchmark_latency.txt
 bench_latency_exit_code=$?
 
-python3 benchmarks/benchmark_throughput.py --input-len 256 --output-len 256 2>&1 | tee benchmark_throughput.txt
+python3 benchmarks/benchmark_throughput.py --input-len 256 --output-len 256 --output-json throughput_results.json 2>&1 | tee benchmark_throughput.txt
 bench_throughput_exit_code=$?
 
 # run server-based benchmarks and upload the result to buildkite
@@ -50,11 +50,16 @@ echo "### Serving Benchmarks" >> benchmark_results.md
 sed -n '1p' benchmark_serving.txt >> benchmark_results.md # first line
 echo "" >> benchmark_results.md
 echo '```' >> benchmark_results.md
-tail -n 20 benchmark_serving.txt >> benchmark_results.md # last 20 lines
+tail -n 24 benchmark_serving.txt >> benchmark_results.md # last 24 lines
 echo '```' >> benchmark_results.md
 
+# if the agent binary is not found, skip uploading the results, exit 0
+if [ ! -f /usr/bin/buildkite-agent ]; then
+    exit 0
+fi
+
 # upload the results to buildkite
-/workspace/buildkite-agent annotate --style "info" --context "benchmark-results" < benchmark_results.md
+buildkite-agent annotate --style "info" --context "benchmark-results" < benchmark_results.md
 
 # exit with the exit code of the benchmarks
 if [ $bench_latency_exit_code -ne 0 ]; then
@@ -69,4 +74,5 @@ if [ $bench_serving_exit_code -ne 0 ]; then
     exit $bench_serving_exit_code
 fi
 
-/workspace/buildkite-agent artifact upload openai-*.json
+rm ShareGPT_V3_unfiltered_cleaned_split.json
+buildkite-agent artifact upload "*.json"

.buildkite/run-cpu-test.sh

@@ -10,5 +10,15 @@ remove_docker_container() { docker rm -f cpu-test || true; }
 trap remove_docker_container EXIT
 remove_docker_container
 
-# Run the image and launch offline inference
+# Run the image
-docker run --network host --env VLLM_CPU_KVCACHE_SPACE=1 --name cpu-test cpu-test python3 examples/offline_inference.py
+docker run -itd -v ~/.cache/huggingface:/root/.cache/huggingface --cpuset-cpus=48-95 --cpuset-mems=1 --network host -e HF_TOKEN --env VLLM_CPU_KVCACHE_SPACE=4 --name cpu-test cpu-test
+
+# offline inference
+docker exec cpu-test bash -c "python3 examples/offline_inference.py"
+
+# Run basic model test
+docker exec cpu-test bash -c "cd tests;
+  pip install pytest Pillow protobuf
+  bash ../.buildkite/download-images.sh
+  cd ../
+  pytest -v -s tests/models --ignore=tests/models/test_llava.py --ignore=tests/models/test_embedding.py --ignore=tests/models/test_registry.py"

.buildkite/run-neuron-test.sh

@@ -4,6 +4,20 @@ set -e
 
 # Try building the docker image
 aws ecr get-login-password --region us-west-2 | docker login --username AWS --password-stdin 763104351884.dkr.ecr.us-west-2.amazonaws.com
+
+# prune old image and containers to save disk space, and only once a day
+# by using a timestamp file in tmp.
+if [ -f /tmp/neuron-docker-build-timestamp ]; then
+    last_build=$(cat /tmp/neuron-docker-build-timestamp)
+    current_time=$(date +%s)
+    if [ $((current_time - last_build)) -gt 86400 ]; then
+        docker system prune -f
+        echo $current_time > /tmp/neuron-docker-build-timestamp
+    fi
+else
+    echo $(date +%s) > /tmp/neuron-docker-build-timestamp
+fi
+
 docker build -t neuron -f Dockerfile.neuron .
 
 # Setup cleanup

.buildkite/test-pipeline.yaml

@@ -5,102 +5,158 @@
 
 steps:
 - label: Regression Test
+  mirror_hardwares: [amd]
   command: pytest -v -s test_regression.py
   working_dir: "/vllm-workspace/tests" # optional
 
 - label: AsyncEngine Test
+  #mirror_hardwares: [amd]
   command: pytest -v -s async_engine
 
 - label: Basic Correctness Test
+  mirror_hardwares: [amd]
   commands:
   - VLLM_ATTENTION_BACKEND=XFORMERS pytest -v -s basic_correctness/test_basic_correctness.py
   - VLLM_ATTENTION_BACKEND=FLASH_ATTN pytest -v -s basic_correctness/test_basic_correctness.py
   - VLLM_ATTENTION_BACKEND=XFORMERS pytest -v -s basic_correctness/test_chunked_prefill.py
   - VLLM_ATTENTION_BACKEND=FLASH_ATTN pytest -v -s basic_correctness/test_chunked_prefill.py
+  - VLLM_TEST_ENABLE_ARTIFICIAL_PREEMPT=1 pytest -v -s basic_correctness/test_preemption.py
 
 - label: Core Test
+  mirror_hardwares: [amd]
   command: pytest -v -s core
 
 - label: Distributed Comm Ops Test
-  command: pytest -v -s test_comm_ops.py
+  #mirror_hardwares: [amd]
-  working_dir: "/vllm-workspace/tests/distributed"
+  command: pytest -v -s distributed/test_comm_ops.py
-  num_gpus: 2 # only support 1 or 2 for now.
+  working_dir: "/vllm-workspace/tests"
+  num_gpus: 2
 
 - label: Distributed Tests
-  working_dir: "/vllm-workspace/tests/distributed"
+  mirror_hardwares: [amd]
-  num_gpus: 2 # only support 1 or 2 for now.
+  working_dir: "/vllm-workspace/tests"
+  num_gpus: 2
   commands:
-  - pytest -v -s test_pynccl.py
+  - VLLM_TEST_SAME_HOST=1 torchrun --nproc-per-node=4 distributed/test_same_node.py
-  - pytest -v -s test_pynccl_library.py
+  - TEST_DIST_MODEL=facebook/opt-125m DISTRIBUTED_EXECUTOR_BACKEND=ray pytest -v -s distributed/test_basic_distributed_correctness.py
-  - TEST_DIST_MODEL=facebook/opt-125m pytest -v -s test_basic_distributed_correctness.py
+  - TEST_DIST_MODEL=meta-llama/Llama-2-7b-hf DISTRIBUTED_EXECUTOR_BACKEND=ray pytest -v -s distributed/test_basic_distributed_correctness.py
-  - TEST_DIST_MODEL=meta-llama/Llama-2-7b-hf pytest -v -s test_basic_distributed_correctness.py
+  - TEST_DIST_MODEL=facebook/opt-125m DISTRIBUTED_EXECUTOR_BACKEND=ray pytest -v -s distributed/test_chunked_prefill_distributed.py
-  - TEST_DIST_MODEL=facebook/opt-125m pytest -v -s test_chunked_prefill_distributed.py
+  - TEST_DIST_MODEL=meta-llama/Llama-2-7b-hf DISTRIBUTED_EXECUTOR_BACKEND=ray pytest -v -s distributed/test_chunked_prefill_distributed.py
-  - TEST_DIST_MODEL=meta-llama/Llama-2-7b-hf pytest -v -s test_chunked_prefill_distributed.py
+  - TEST_DIST_MODEL=facebook/opt-125m DISTRIBUTED_EXECUTOR_BACKEND=mp pytest -v -s distributed/test_basic_distributed_correctness.py
+  - TEST_DIST_MODEL=meta-llama/Llama-2-7b-hf DISTRIBUTED_EXECUTOR_BACKEND=mp pytest -v -s distributed/test_basic_distributed_correctness.py
+  - TEST_DIST_MODEL=facebook/opt-125m DISTRIBUTED_EXECUTOR_BACKEND=mp pytest -v -s distributed/test_chunked_prefill_distributed.py
+  - TEST_DIST_MODEL=meta-llama/Llama-2-7b-hf DISTRIBUTED_EXECUTOR_BACKEND=mp pytest -v -s distributed/test_chunked_prefill_distributed.py
+  - pytest -v -s spec_decode/e2e/test_integration_dist.py
+  - CUDA_VISIBLE_DEVICES=0,1 pytest -v -s test_sharded_state_loader.py
+
+- label: Distributed Tests (Multiple Groups)
+  #mirror_hardwares: [amd]
+  working_dir: "/vllm-workspace/tests"
+  num_gpus: 4
+  commands:
+  - pytest -v -s distributed/test_pynccl.py
 
 - label: Engine Test
+  mirror_hardwares: [amd]
   command: pytest -v -s engine tokenization test_sequence.py test_config.py test_logger.py
 
 - label: Entrypoints Test
+  mirror_hardwares: [amd]
+
   commands:
-  # these tests have to be separated, because each one will allocate all posible GPU memory
+  - pytest -v -s entrypoints -m llm
-  - pytest -v -s entrypoints --ignore=entrypoints/test_server_oot_registration.py
+  - pytest -v -s entrypoints -m openai
-  - pytest -v -s entrypoints/test_server_oot_registration.py
 
 - label: Examples Test
   working_dir: "/vllm-workspace/examples"
+  mirror_hardwares: [amd]
   commands:
     # install aws cli for llava_example.py
-    - pip install awscli
+    # install tensorizer for tensorize_vllm_model.py
+    - pip install awscli tensorizer
     - python3 offline_inference.py
     - python3 offline_inference_with_prefix.py
     - python3 llm_engine_example.py
     - python3 llava_example.py
+    - python3 tensorize_vllm_model.py --model facebook/opt-125m serialize --serialized-directory /tmp/ --suffix v1 && python3 tensorize_vllm_model.py --model facebook/opt-125m deserialize --path-to-tensors /tmp/vllm/facebook/opt-125m/v1/model.tensors
+
+- label: Inputs Test
+  #mirror_hardwares: [amd]
+  commands:
+    - bash ../.buildkite/download-images.sh
+    - pytest -v -s test_inputs.py
+    - pytest -v -s multimodal
 
 - label: Kernels Test %N
+  #mirror_hardwares: [amd]
   command: pytest -v -s kernels --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
   parallelism: 4
 
 - label: Models Test
+  #mirror_hardwares: [amd]
   commands:
-    - bash ../.buildkite/download-images.sh
+    - pytest -v -s models -m \"not llava\"
-    - pytest -v -s models --ignore=models/test_llava.py --ignore=models/test_mistral.py
 
 - label: Llava Test
+  mirror_hardwares: [amd]
   commands:
     - bash ../.buildkite/download-images.sh
-    - pytest -v -s models/test_llava.py
+    - pytest -v -s models -m llava
 
 - label: Prefix Caching Test
+  mirror_hardwares: [amd]
   commands:
     - pytest -v -s prefix_caching
 
 - label: Samplers Test
+  #mirror_hardwares: [amd]
   command: pytest -v -s samplers
 
 - label: LogitsProcessor Test
+  mirror_hardwares: [amd]
   command: pytest -v -s test_logits_processor.py
 
+- label: Utils Test
+  command: pytest -v -s test_utils.py
+
 - label: Worker Test
+  mirror_hardwares: [amd]
   command: pytest -v -s worker
 
 - label: Speculative decoding tests
-  command: pytest -v -s spec_decode
+  #mirror_hardwares: [amd]
+  commands:
+    # See https://github.com/vllm-project/vllm/issues/5152
+    - export VLLM_ATTENTION_BACKEND=XFORMERS
+    - pytest -v -s spec_decode
 
 - label: LoRA Test %N
-  command: pytest -v -s lora --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT
+  #mirror_hardwares: [amd]
+  command: pytest -v -s lora --shard-id=$$BUILDKITE_PARALLEL_JOB --num-shards=$$BUILDKITE_PARALLEL_JOB_COUNT --ignore=lora/test_long_context.py
   parallelism: 4
 
+- label: LoRA Long Context (Distributed)
+  #mirror_hardwares: [amd]
+  num_gpus: 4
+  # This test runs llama 13B, so it is required to run on 4 GPUs.
+  commands:
+    - pytest -v -s -x lora/test_long_context.py
+
 - label: Tensorizer Test
+  #mirror_hardwares: [amd]
   command: apt-get install curl libsodium23 && pytest -v -s tensorizer_loader
 
 - label: Metrics Test
+  mirror_hardwares: [amd]
   command: pytest -v -s metrics
 
 - label: Quantization Test
+  #mirror_hardwares: [amd]
   command: pytest -v -s quantization
 
 - label: Benchmarks
   working_dir: "/vllm-workspace/.buildkite"
+  mirror_hardwares: [amd]
   commands:
   - pip install aiohttp
   - bash run-benchmarks.sh

.buildkite/test-template-aws.j2

@@ -0,0 +1,64 @@
+{% set docker_image = "public.ecr.aws/q9t5s3a7/vllm-ci-test-repo:$BUILDKITE_COMMIT" %}
+{% set default_working_dir = "/vllm-workspace/tests" %}
+
+steps:
+  - label: ":docker: build image"
+    agents:
+      queue: cpu_queue
+    commands:
+      - "aws ecr-public get-login-password --region us-east-1 | docker login --username AWS --password-stdin public.ecr.aws/q9t5s3a7"
+      - "docker build --build-arg max_jobs=16 --tag {{ docker_image }} --target test --progress plain ."
+      - "docker push {{ docker_image }}"
+    env:
+      DOCKER_BUILDKIT: "1"
+    retry:
+      automatic:
+        - exit_status: -1  # Agent was lost
+          limit: 5
+        - exit_status: -10  # Agent was lost
+          limit: 5
+  - wait
+
+  {% for step in steps %}
+  - label: "{{ step.label }}"
+    agents:
+      {% if step.label == "Documentation Build" %}
+      queue: small_cpu_queue
+      {% elif step.no_gpu %}
+      queue: cpu_queue
+      {% elif step.num_gpus == 2 or step.num_gpus == 4 %}
+      queue: gpu_4_queue
+      {% else %}
+      queue: gpu_1_queue
+      {% endif %}
+    soft_fail: true
+    {% if step.parallelism %}
+    parallelism: {{ step.parallelism }}
+    {% endif %}
+    retry:
+      automatic:
+        - exit_status: -1  # Agent was lost
+          limit: 5
+        - exit_status: -10  # Agent was lost
+          limit: 5
+    plugins:
+      - docker#v5.2.0:
+          image: {{ docker_image }}
+          always-pull: true
+          propagate-environment: true
+          {% if not step.no_gpu %}
+          gpus: all
+          {% endif %}
+          {% if step.label == "Benchmarks" %}
+          mount-buildkite-agent: true
+          {% endif %}
+          command: ["bash", "-c", "cd {{ (step.working_dir or default_working_dir) | safe  }} && {{ step.command  or (step.commands | join(' && ')) | safe }}"]
+          environment:
+            - VLLM_USAGE_SOURCE=ci-test
+            - HF_TOKEN
+            {% if step.label == "Speculative decoding tests" %}
+            - VLLM_ATTENTION_BACKEND=XFORMERS
+            {% endif %}
+          volumes:
+            - /dev/shm:/dev/shm
+  {% endfor %}

.buildkite/test-template.j2

@@ -3,7 +3,6 @@
 {% set default_working_dir = "/vllm-workspace/tests" %}
 
 steps:
-
   - label: ":docker: build image"
     commands:
       - "docker build --build-arg max_jobs=16 --tag {{ docker_image }} --target test --progress plain ."
@@ -14,19 +13,36 @@ steps:
       automatic:
         - exit_status: -1  # Agent was lost
           limit: 5
+        - exit_status: -10  # Agent was lost
+          limit: 5
   - wait
 
-  - label: "AMD Test"
+  - group: "AMD Tests"
+    depends_on: ~
+    steps:
+    {% for step in steps %}
+    {% if step.mirror_hardwares and "amd" in step.mirror_hardwares %}
+      - label: "AMD: {{ step.label }}"
         agents:
           queue: amd
-    command: bash .buildkite/run-amd-test.sh
+        command: bash .buildkite/run-amd-test.sh "cd {{ (step.working_dir or default_working_dir) | safe  }} ; {{ step.command  or (step.commands | join(" ; ")) | safe }}"
+        env:
+          DOCKER_BUILDKIT: "1"
+        soft_fail: true
+    {% endif %}
+    {% endfor %}
 
   - label: "Neuron Test"
+    depends_on: ~
     agents:
       queue: neuron
     command: bash .buildkite/run-neuron-test.sh
+    soft_fail: false
 
-  - label: "CPU Test"
+  - label: "Intel Test"
+    depends_on: ~
+    agents:
+      queue: intel
     command: bash .buildkite/run-cpu-test.sh
 
   {% for step in steps %}
@@ -41,9 +57,14 @@ steps:
       automatic:
         - exit_status: -1  # Agent was lost
           limit: 5
+        - exit_status: -10  # Agent was lost
+          limit: 5
     plugins:
       - kubernetes:
           podSpec:
+            {% if step.num_gpus %}
+            priorityClassName: gpu-priority-cls-{{ step.num_gpus }}
+            {% endif %}
             volumes:
               - name: dshm
                 emptyDir:

.clang-format

@@ -0,0 +1,26 @@
+BasedOnStyle: Google
+UseTab: Never
+IndentWidth: 2
+ColumnLimit: 80
+
+# Force pointers to the type for C++.
+DerivePointerAlignment: false
+PointerAlignment: Left
+
+# Reordering #include statements can (and currently will) introduce errors
+SortIncludes: false
+
+# Style choices
+AlignConsecutiveAssignments: false
+AlignConsecutiveDeclarations: false
+IndentPPDirectives: BeforeHash
+
+IncludeCategories:
+  - Regex:           '^<'
+    Priority:        4
+  - Regex:           '^"(llvm|llvm-c|clang|clang-c|mlir|mlir-c)/'
+    Priority:        3
+  - Regex:           '^"(qoda|\.\.)/'
+    Priority:        2
+  - Regex:           '.*'
+    Priority:        1

.github/ISSUE_TEMPLATE/400-bug report.yml

@@ -59,6 +59,8 @@ body:
 
       Please also paste or describe the results you observe instead of the expected results. If you observe an error, please paste the error message including the **full** traceback of the exception. It may be relevant to wrap error messages in ```` ```triple quotes blocks``` ````.
 
+      Please set the environment variable `export VLLM_LOGGING_LEVEL=DEBUG` to turn on more logging to help debugging potential issues.
+
       If you experienced crashes or hangs, it would be helpful to run vllm with `export VLLM_TRACE_FUNCTION=1` . All the function calls in vllm will be recorded. Inspect these log files, and tell which function crashes or hangs.
     placeholder: |
       A clear and concise description of what the bug is.

.github/ISSUE_TEMPLATE/750-RFC.yml

@@ -0,0 +1,49 @@
+name: 💬 Request for comments (RFC).
+description: Ask for feedback on major architectural changes or design choices.
+title: "[RFC]: "
+labels: ["RFC"]
+
+body:
+- type: markdown
+  attributes:
+    value: >
+      #### Please take a look at previous [RFCs](https://github.com/vllm-project/vllm/issues?q=label%3ARFC+sort%3Aupdated-desc) for reference.
+- type: textarea
+  attributes:
+    label: Motivation.
+    description: >
+      The motivation of the RFC.
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: Proposed Change.
+    description: >
+      The proposed change of the RFC.
+  validations:
+    required: true
+- type: textarea
+  attributes:
+    label: Feedback Period.
+    description: >
+      The feedback period of the RFC. Usually at least one week.
+  validations:
+    required: false
+- type: textarea
+  attributes:
+    label: CC List.
+    description: >
+      The list of people you want to CC.
+  validations:
+    required: false
+- type: textarea
+  attributes:
+    label: Any Other Things.
+    description: >
+      Any other things you would like to mention.
+  validations:
+    required: false
+- type: markdown
+  attributes:
+    value: >
+      Thanks for contributing 🎉!

.github/workflows/clang-format.yml

@@ -0,0 +1,42 @@
+name: clang-format
+
+on:
+  # Trigger the workflow on push or pull request,
+  # but only for the main branch
+  push:
+    branches:
+      - main
+  pull_request:
+    branches:
+      - main
+
+jobs:
+  clang-format:
+    runs-on: ubuntu-latest
+    strategy:
+      matrix:
+        python-version: ["3.11"]
+    steps:
+    - uses: actions/checkout@v2
+    - name: Set up Python ${{ matrix.python-version }}
+      uses: actions/setup-python@v2
+      with:
+        python-version: ${{ matrix.python-version }}
+    - name: Install dependencies
+      run: |
+        python -m pip install --upgrade pip
+        pip install clang-format==18.1.5
+    - name: Running clang-format
+      run: |
+        EXCLUDES=(
+            'csrc/moe/topk_softmax_kernels.cu'
+            'csrc/punica/bgmv/bgmv_bf16_bf16_bf16.cu'
+            'csrc/punica/bgmv/bgmv_config.h'
+            'csrc/punica/bgmv/bgmv_impl.cuh'
+            'csrc/punica/bgmv/vec_dtypes.cuh'
+            'csrc/punica/punica_ops.cu'
+            'csrc/punica/type_convert.h'
+        )
+        find csrc/ \( -name '*.h' -o -name '*.cpp' -o -name '*.cu' -o -name '*.cuh' \) -print \
+            | grep -vFf <(printf "%s\n" "${EXCLUDES[@]}") \
+            | xargs clang-format --dry-run --Werror

.github/workflows/mypy.yaml

@@ -33,19 +33,19 @@ jobs:
     - name: Mypy
       run: |
         mypy vllm/attention --config-file pyproject.toml
-        # TODO(sang): Fix nested dir
+        mypy vllm/core --config-file pyproject.toml
-        mypy vllm/core/*.py --follow-imports=skip --config-file pyproject.toml
         mypy vllm/distributed --config-file pyproject.toml
         mypy vllm/entrypoints --config-file pyproject.toml
         mypy vllm/executor --config-file pyproject.toml
+        mypy vllm/multimodal --config-file pyproject.toml
         mypy vllm/usage --config-file pyproject.toml
         mypy vllm/*.py --config-file pyproject.toml
         mypy vllm/transformers_utils --config-file pyproject.toml
         mypy vllm/engine  --config-file pyproject.toml
         mypy vllm/worker --config-file pyproject.toml
         mypy vllm/spec_decode --config-file pyproject.toml
-        # TODO(sang): Fix nested dir
+        mypy vllm/model_executor  --config-file pyproject.toml
-        mypy vllm/model_executor/*.py  --config-file pyproject.toml
+        mypy vllm/lora --config-file pyproject.toml
-        # TODO(sang): Fix nested dir
+        mypy vllm/logging --config-file pyproject.toml
-        # mypy vllm/lora/*.py --config-file pyproject.toml
+        mypy vllm/model_executor --config-file pyproject.toml
 

.github/workflows/publish.yml

@@ -49,7 +49,7 @@ jobs:
       matrix:
           os: ['ubuntu-20.04']
           python-version: ['3.8', '3.9', '3.10', '3.11']
-          pytorch-version: ['2.2.1']  # Must be the most recent version that meets requirements-cuda.txt.
+          pytorch-version: ['2.3.0']  # Must be the most recent version that meets requirements-cuda.txt.
           cuda-version: ['11.8', '12.1']
 
     steps:
@@ -58,6 +58,9 @@ jobs:
 
       - name: Setup ccache
         uses: hendrikmuhs/ccache-action@v1.2
+        with:
+          create-symlink: true
+          key: ${{ github.job }}-${{ matrix.python-version }}-${{ matrix.cuda-version }}
 
       - name: Set up Linux Env
         if: ${{ runner.os == 'Linux' }}
@@ -79,6 +82,8 @@ jobs:
 
       - name: Build wheel
         shell: bash
+        env:
+          CMAKE_BUILD_TYPE: Release # do not compile with debug symbol to reduce wheel size
         run: |
           bash -x .github/workflows/scripts/build.sh ${{ matrix.python-version }} ${{ matrix.cuda-version }}
           wheel_name=$(ls dist/*whl | xargs -n 1 basename)

.github/workflows/scripts/create_release.js

@@ -8,7 +8,7 @@ module.exports = async (github, context, core) => {
 			generate_release_notes: true,
 			name: process.env.RELEASE_TAG,
 			owner: context.repo.owner,
-			prerelease: false,
+			prerelease: true,
 			repo: context.repo.repo,
 			tag_name: process.env.RELEASE_TAG,
 		});

CMakeLists.txt

@@ -31,7 +31,7 @@ set(HIP_SUPPORTED_ARCHS "gfx906;gfx908;gfx90a;gfx940;gfx941;gfx942;gfx1030;gfx11
 # requirements.txt files and should be kept consistent.  The ROCm torch
 # versions are derived from Dockerfile.rocm
 #
-set(TORCH_SUPPORTED_VERSION_CUDA "2.2.1")
+set(TORCH_SUPPORTED_VERSION_CUDA "2.3.0")
 set(TORCH_SUPPORTED_VERSION_ROCM_5X "2.0.1")
 set(TORCH_SUPPORTED_VERSION_ROCM_6X "2.1.1")
 
@@ -66,19 +66,6 @@ endif()
 #
 find_package(Torch REQUIRED)
 
-#
-# Normally `torch.utils.cpp_extension.CUDAExtension` would add
-# `libtorch_python.so` for linking against an extension. Torch's cmake
-# configuration does not include this library (presumably since the cmake
-# config is used for standalone C++ binaries that link against torch).
-# The `libtorch_python.so` library defines some of the glue code between
-# torch/python via pybind and is required by VLLM extensions for this
-# reason. So, add it by manually with `find_library` using torch's
-# installed library path.
-#
-find_library(torch_python_LIBRARY torch_python PATHS
-  "${TORCH_INSTALL_PREFIX}/lib")
-
 #
 # Forward the non-CUDA device extensions to external CMake scripts.
 #
@@ -167,17 +154,47 @@ set(VLLM_EXT_SRC
   "csrc/layernorm_kernels.cu"
   "csrc/quantization/squeezellm/quant_cuda_kernel.cu"
   "csrc/quantization/gptq/q_gemm.cu"
-  "csrc/quantization/fp8/fp8_cuda_kernels.cu"
+  "csrc/quantization/compressed_tensors/int8_quant_kernels.cu"
+  "csrc/quantization/fp8/common.cu"
   "csrc/cuda_utils_kernels.cu"
   "csrc/moe_align_block_size_kernels.cu"
-  "csrc/pybind.cpp")
+  "csrc/torch_bindings.cpp")
 
 if(VLLM_GPU_LANG STREQUAL "CUDA")
+  include(FetchContent)
+  SET(CUTLASS_ENABLE_HEADERS_ONLY=ON)
+  FetchContent_Declare(
+        cutlass
+        GIT_REPOSITORY https://github.com/nvidia/cutlass.git
+        # CUTLASS 3.5.0
+        GIT_TAG 7d49e6c7e2f8896c47f586706e67e1fb215529dc
+  )
+  FetchContent_MakeAvailable(cutlass)
+
   list(APPEND VLLM_EXT_SRC
     "csrc/quantization/aqlm/gemm_kernels.cu"
     "csrc/quantization/awq/gemm_kernels.cu"
-    "csrc/quantization/marlin/marlin_cuda_kernel.cu"
+    "csrc/quantization/marlin/dense/marlin_cuda_kernel.cu"
-    "csrc/custom_all_reduce.cu")
+    "csrc/quantization/marlin/sparse/marlin_24_cuda_kernel.cu"
+    "csrc/quantization/gptq_marlin/gptq_marlin.cu"
+    "csrc/quantization/gptq_marlin/gptq_marlin_repack.cu"
+    "csrc/custom_all_reduce.cu"
+    "csrc/quantization/cutlass_w8a8/scaled_mm_dq_entry.cu"
+    "csrc/quantization/cutlass_w8a8/scaled_mm_dq_c2x.cu"
+    "csrc/quantization/cutlass_w8a8/scaled_mm_dq_c3x.cu")
+
+  #
+  # The CUTLASS kernels for Hopper require sm90a to be enabled.
+  # This is done via the below gencode option, BUT that creates kernels for both sm90 and sm90a.
+  # That adds an extra 17MB to compiled binary, so instead we selectively enable it.
+  if(${CMAKE_CUDA_COMPILER_VERSION} VERSION_GREATER 12.0)
+    set_source_files_properties(
+          "csrc/quantization/cutlass_w8a8/scaled_mm_dq_c3x.cu"
+          PROPERTIES
+          COMPILE_FLAGS
+          "-gencode arch=compute_90a,code=sm_90a")
+  endif()
+
 endif()
 
 define_gpu_extension_target(
@@ -187,6 +204,8 @@ define_gpu_extension_target(
   SOURCES ${VLLM_EXT_SRC}
   COMPILE_FLAGS ${VLLM_GPU_FLAGS}
   ARCHITECTURES ${VLLM_GPU_ARCHES}
+  INCLUDE_DIRECTORIES ${CUTLASS_INCLUDE_DIR};${CUTLASS_TOOLS_UTIL_INCLUDE_DIR}
+  USE_SABI 3
   WITH_SOABI)
 
 #
@@ -194,7 +213,7 @@ define_gpu_extension_target(
 #
 
 set(VLLM_MOE_EXT_SRC
-  "csrc/moe/moe_ops.cpp"
+  "csrc/moe/torch_bindings.cpp"
   "csrc/moe/topk_softmax_kernels.cu")
 
 define_gpu_extension_target(
@@ -204,6 +223,7 @@ define_gpu_extension_target(
   SOURCES ${VLLM_MOE_EXT_SRC}
   COMPILE_FLAGS ${VLLM_GPU_FLAGS}
   ARCHITECTURES ${VLLM_GPU_ARCHES}
+  USE_SABI 3
   WITH_SOABI)
 
 #
@@ -217,7 +237,8 @@ set(VLLM_PUNICA_EXT_SRC
   "csrc/punica/bgmv/bgmv_fp16_fp32_fp16.cu"
   "csrc/punica/bgmv/bgmv_fp32_bf16_bf16.cu"
   "csrc/punica/bgmv/bgmv_fp32_fp16_fp16.cu"
-  "csrc/punica/punica_ops.cc")
+  "csrc/punica/punica_ops.cu"
+  "csrc/punica/torch_bindings.cpp")
 
 #
 # Copy GPU compilation flags+update for punica
@@ -241,6 +262,9 @@ if (${VLLM_GPU_LANG} STREQUAL "CUDA")
     endif()
   endforeach()
   message(STATUS "Punica target arches: ${VLLM_PUNICA_GPU_ARCHES}")
+elseif(${VLLM_GPU_LANG} STREQUAL "HIP")
+  set(VLLM_PUNICA_GPU_ARCHES ${VLLM_GPU_ARCHES})
+  message(STATUS "Punica target arches: ${VLLM_PUNICA_GPU_ARCHES}")
 endif()
 
 if (VLLM_PUNICA_GPU_ARCHES)
@@ -251,6 +275,7 @@ if (VLLM_PUNICA_GPU_ARCHES)
     SOURCES ${VLLM_PUNICA_EXT_SRC}
     COMPILE_FLAGS ${VLLM_PUNICA_GPU_FLAGS}
     ARCHITECTURES ${VLLM_PUNICA_GPU_ARCHES}
+    USE_SABI 3
     WITH_SOABI)
 else()
   message(WARNING "Unable to create _punica_C target because none of the "
@@ -275,9 +300,7 @@ add_custom_target(default)
 if(VLLM_GPU_LANG STREQUAL "CUDA" OR VLLM_GPU_LANG STREQUAL "HIP")
   message(STATUS "Enabling C extension.")
   add_dependencies(default _C)
-endif()
 
-if(VLLM_GPU_LANG STREQUAL "CUDA")
   message(STATUS "Enabling moe extension.")
   add_dependencies(default _moe_C)
 

Dockerfile

@@ -1,9 +1,13 @@
 # The vLLM Dockerfile is used to construct vLLM image that can be directly used
 # to run the OpenAI compatible server.
 
+# Please update any changes made here to
+# docs/source/dev/dockerfile/dockerfile.rst and
+# docs/source/assets/dev/dockerfile-stages-dependency.png
+
 #################### BASE BUILD IMAGE ####################
 # prepare basic build environment
-FROM nvidia/cuda:12.1.0-devel-ubuntu22.04 AS dev
+FROM nvidia/cuda:12.4.1-devel-ubuntu22.04 AS dev
 
 RUN apt-get update -y \
     && apt-get install -y python3-pip git
@@ -12,7 +16,7 @@ RUN apt-get update -y \
 # https://github.com/pytorch/pytorch/issues/107960 -- hopefully
 # this won't be needed for future versions of this docker image
 # or future versions of triton.
-RUN ldconfig /usr/local/cuda-12.1/compat/
+RUN ldconfig /usr/local/cuda-12.4/compat/
 
 WORKDIR /workspace
 
@@ -71,34 +75,15 @@ RUN --mount=type=cache,target=/root/.cache/ccache \
     --mount=type=cache,target=/root/.cache/pip \
     python3 setup.py bdist_wheel --dist-dir=dist
 
-# the `vllm_nccl` package must be installed from source distribution
+# check the size of the wheel, we cannot upload wheels larger than 100MB
-# pip is too smart to store a wheel in the cache, and other CI jobs
+COPY .buildkite/check-wheel-size.py check-wheel-size.py
-# will directly use the wheel from the cache, which is not what we want.
+RUN python3 check-wheel-size.py dist
-# we need to remove it manually
+
-RUN --mount=type=cache,target=/root/.cache/pip \
-    pip cache remove vllm_nccl*
 #################### EXTENSION Build IMAGE ####################
 
-#################### FLASH_ATTENTION Build IMAGE ####################
-FROM dev as flash-attn-builder
-# max jobs used for build
-ARG max_jobs=2
-ENV MAX_JOBS=${max_jobs}
-# flash attention version
-ARG flash_attn_version=v2.5.6
-ENV FLASH_ATTN_VERSION=${flash_attn_version}
-
-WORKDIR /usr/src/flash-attention-v2
-
-# Download the wheel or build it if a pre-compiled release doesn't exist
-RUN pip --verbose wheel flash-attn==${FLASH_ATTN_VERSION} \
-    --no-build-isolation --no-deps --no-cache-dir
-
-#################### FLASH_ATTENTION Build IMAGE ####################
-
 #################### vLLM installation IMAGE ####################
 # image with vLLM installed
-FROM nvidia/cuda:12.1.0-base-ubuntu22.04 AS vllm-base
+FROM nvidia/cuda:12.4.1-base-ubuntu22.04 AS vllm-base
 WORKDIR /vllm-workspace
 
 RUN apt-get update -y \
@@ -108,16 +93,12 @@ RUN apt-get update -y \
 # https://github.com/pytorch/pytorch/issues/107960 -- hopefully
 # this won't be needed for future versions of this docker image
 # or future versions of triton.
-RUN ldconfig /usr/local/cuda-12.1/compat/
+RUN ldconfig /usr/local/cuda-12.4/compat/
 
 # install vllm wheel first, so that torch etc will be installed
 RUN --mount=type=bind,from=build,src=/workspace/dist,target=/vllm-workspace/dist \
     --mount=type=cache,target=/root/.cache/pip \
     pip install dist/*.whl --verbose
-
-RUN --mount=type=bind,from=flash-attn-builder,src=/usr/src/flash-attention-v2,target=/usr/src/flash-attention-v2 \
-    --mount=type=cache,target=/root/.cache/pip \
-    pip install /usr/src/flash-attention-v2/*.whl --no-cache-dir
 #################### vLLM installation IMAGE ####################
 
 

Dockerfile.cpu

@@ -1,13 +1,15 @@
 # This vLLM Dockerfile is used to construct image that can build and run vLLM on x86 CPU platform.
 
-FROM ubuntu:22.04
+FROM ubuntu:22.04 AS cpu-test-1
 
 RUN apt-get update  -y \
     && apt-get install -y git wget vim numactl gcc-12 g++-12 python3 python3-pip \
     && update-alternatives --install /usr/bin/gcc gcc /usr/bin/gcc-12 10 --slave /usr/bin/g++ g++ /usr/bin/g++-12
 
 RUN pip install --upgrade pip \
-    && pip install wheel packaging ninja setuptools>=49.4.0 numpy
+    && pip install wheel packaging ninja "setuptools>=49.4.0" numpy
+
+FROM cpu-test-1 AS build
 
 COPY ./ /workspace/vllm
 
@@ -17,4 +19,8 @@ RUN pip install -v -r requirements-cpu.txt --extra-index-url https://download.py
 
 RUN VLLM_TARGET_DEVICE=cpu python3 setup.py install
 
+WORKDIR /workspace/
+
+RUN ln -s /workspace/vllm/tests  && ln -s /workspace/vllm/examples && ln -s /workspace/vllm/benchmarks
+
 CMD ["/bin/bash"]

Dockerfile.neuron

@@ -28,7 +28,7 @@ COPY ./requirements-neuron.txt /app/vllm/requirements-neuron.txt
 RUN cd /app/vllm \
     && python3 -m pip install -U -r requirements-neuron.txt
 
-ENV VLLM_BUILD_WITH_NEURON 1
+ENV VLLM_TARGET_DEVICE neuron
 RUN cd /app/vllm \
     && pip install -e . \
     && cd ..

Dockerfile.rocm

@@ -46,7 +46,7 @@ RUN apt-get update && apt-get install -y \
 
 ### Mount Point ###
 # When launching the container, mount the code directory to /app
-ARG APP_MOUNT=/app
+ARG APP_MOUNT=/vllm-workspace
 VOLUME [ ${APP_MOUNT} ]
 WORKDIR ${APP_MOUNT}
 
@@ -89,18 +89,27 @@ RUN if [ "$BUILD_TRITON" = "1" ]; then \
     && cd ../..; \
     fi
 
-COPY ./ /app/vllm
+WORKDIR /vllm-workspace
+COPY . .
 
+#RUN python3 -m pip install pynvml # to be removed eventually
 RUN python3 -m pip install --upgrade pip numba
 
-RUN cd /app \
+# make sure punica kernels are built (for LoRA)
-    && cd vllm \
+ENV VLLM_INSTALL_PUNICA_KERNELS=1
-    && pip install -U -r requirements-rocm.txt \
+# Workaround for ray >= 2.10.0
-    && patch /opt/rocm/include/hip/amd_detail/amd_hip_bf16.h /app/vllm/rocm_patch/rocm_bf16.patch \
+ENV RAY_EXPERIMENTAL_NOSET_ROCR_VISIBLE_DEVICES=1
+
+ENV VLLM_NCCL_SO_PATH=/opt/rocm/lib/librccl.so
+
+RUN --mount=type=cache,target=/root/.cache/pip \
+    pip install -U -r requirements-rocm.txt \
+    && patch /opt/rocm/include/hip/amd_detail/amd_hip_bf16.h ./rocm_patch/rocm_bf16.patch \
     && python3 setup.py install \
+    && cp build/lib.linux-x86_64-cpython-39/vllm/_C.abi3.so vllm/ \
+    && cp build/lib.linux-x86_64-cpython-39/vllm/_punica_C.abi3.so vllm/ \
+    && cp build/lib.linux-x86_64-cpython-39/vllm/_moe_C.abi3.so vllm/ \
     && cd ..
 
-RUN python3 -m pip install --upgrade pip
-RUN python3 -m pip install --no-cache-dir ray[all]==2.9.3
 
 CMD ["/bin/bash"]

MANIFEST.in

@@ -1,6 +1,9 @@
 include LICENSE
 include requirements-common.txt
 include requirements-cuda.txt
+include requirements-rocm.txt
+include requirements-neuron.txt
+include requirements-cpu.txt
 include CMakeLists.txt
 
 recursive-include cmake *

README.md

@@ -14,6 +14,24 @@ Easy, fast, and cheap LLM serving for everyone
 
 </p>
 
+---
+
+**Ray Summit CPF is Open (June 4th to June 20th)!**
+
+There will be a track for vLLM at the Ray Summit (09/30-10/02, SF) this year!
+If you have cool projects related to vLLM or LLM inference, we would love to see your proposals.
+This will be a great chance for everyone in the community to get together and learn.
+Please submit your proposal [here](https://raysummit.anyscale.com/flow/anyscale/raysummit2024/landing/page/eventsite)
+
+**The Fourth vLLM Bay Area Meetup (June 11th 5:30pm-8pm PT)**
+
+We are thrilled to announce our fourth vLLM Meetup!
+The vLLM team will share recent updates and roadmap.
+We will also have vLLM collaborators from BentoML and Cloudflare coming up to the stage to discuss their experience in deploying LLMs with vLLM.
+Please register [here](https://lu.ma/agivllm) and join us!
+
+---
+
 *Latest News* 🔥
 - [2024/04] We hosted [the third vLLM meetup](https://robloxandvllmmeetup2024.splashthat.com/) with Roblox! Please find the meetup slides [here](https://docs.google.com/presentation/d/1A--47JAK4BJ39t954HyTkvtfwn0fkqtsL8NGFuslReM/edit?usp=sharing).
 - [2024/01] We hosted [the second vLLM meetup](https://lu.ma/ygxbpzhl) in SF! Please find the meetup slides [here](https://docs.google.com/presentation/d/12mI2sKABnUw5RBWXDYY-HtHth4iMSNcEoQ10jDQbxgA/edit?usp=sharing).
@@ -51,40 +69,14 @@ vLLM is flexible and easy to use with:
 - (Experimental) Prefix caching support
 - (Experimental) Multi-lora support
 
-vLLM seamlessly supports many Hugging Face models, including the following architectures:
+vLLM seamlessly supports most popular open-source models on HuggingFace, including:
+- Transformer-like LLMs (e.g., Llama)
+- Mixture-of-Expert LLMs (e.g., Mixtral)
+- Multi-modal LLMs (e.g., LLaVA)
 
-- Aquila & Aquila2 (`BAAI/AquilaChat2-7B`, `BAAI/AquilaChat2-34B`, `BAAI/Aquila-7B`, `BAAI/AquilaChat-7B`, etc.)
+Find the full list of supported models [here](https://docs.vllm.ai/en/latest/models/supported_models.html).
-- Baichuan & Baichuan2 (`baichuan-inc/Baichuan2-13B-Chat`, `baichuan-inc/Baichuan-7B`, etc.)
+
-- BLOOM (`bigscience/bloom`, `bigscience/bloomz`, etc.)
+## Getting Started
-- ChatGLM (`THUDM/chatglm2-6b`, `THUDM/chatglm3-6b`, etc.)
-- Command-R (`CohereForAI/c4ai-command-r-v01`, etc.)
-- DBRX (`databricks/dbrx-base`, `databricks/dbrx-instruct` etc.)
-- DeciLM (`Deci/DeciLM-7B`, `Deci/DeciLM-7B-instruct`, etc.)
-- Falcon (`tiiuae/falcon-7b`, `tiiuae/falcon-40b`, `tiiuae/falcon-rw-7b`, etc.)
-- Gemma (`google/gemma-2b`, `google/gemma-7b`, etc.)
-- GPT-2 (`gpt2`, `gpt2-xl`, etc.)
-- GPT BigCode (`bigcode/starcoder`, `bigcode/gpt_bigcode-santacoder`, etc.)
-- GPT-J (`EleutherAI/gpt-j-6b`, `nomic-ai/gpt4all-j`, etc.)
-- GPT-NeoX (`EleutherAI/gpt-neox-20b`, `databricks/dolly-v2-12b`, `stabilityai/stablelm-tuned-alpha-7b`, etc.)
-- InternLM (`internlm/internlm-7b`, `internlm/internlm-chat-7b`, etc.)
-- InternLM2 (`internlm/internlm2-7b`, `internlm/internlm2-chat-7b`, etc.)
-- Jais (`core42/jais-13b`, `core42/jais-13b-chat`, `core42/jais-30b-v3`, `core42/jais-30b-chat-v3`, etc.)
-- LLaMA, Llama 2, and Meta Llama 3 (`meta-llama/Meta-Llama-3-8B-Instruct`, `meta-llama/Meta-Llama-3-70B-Instruct`, `meta-llama/Llama-2-70b-hf`, `lmsys/vicuna-13b-v1.3`, `young-geng/koala`, `openlm-research/open_llama_13b`, etc.)
-- MiniCPM (`openbmb/MiniCPM-2B-sft-bf16`, `openbmb/MiniCPM-2B-dpo-bf16`, etc.)
-- Mistral (`mistralai/Mistral-7B-v0.1`, `mistralai/Mistral-7B-Instruct-v0.1`, etc.)
-- Mixtral (`mistralai/Mixtral-8x7B-v0.1`, `mistralai/Mixtral-8x7B-Instruct-v0.1`, `mistral-community/Mixtral-8x22B-v0.1`, etc.)
-- MPT (`mosaicml/mpt-7b`, `mosaicml/mpt-30b`, etc.)
-- OLMo (`allenai/OLMo-1B`, `allenai/OLMo-7B`, etc.)
-- OPT (`facebook/opt-66b`, `facebook/opt-iml-max-30b`, etc.)
-- Orion (`OrionStarAI/Orion-14B-Base`, `OrionStarAI/Orion-14B-Chat`, etc.)
-- Phi (`microsoft/phi-1_5`, `microsoft/phi-2`, etc.)
-- Qwen (`Qwen/Qwen-7B`, `Qwen/Qwen-7B-Chat`, etc.)
-- Qwen2 (`Qwen/Qwen1.5-7B`, `Qwen/Qwen1.5-7B-Chat`, etc.)
-- Qwen2MoE (`Qwen/Qwen1.5-MoE-A2.7B`, `Qwen/Qwen1.5-MoE-A2.7B-Chat`, etc.)
-- StableLM(`stabilityai/stablelm-3b-4e1t`, `stabilityai/stablelm-base-alpha-7b-v2`, etc.)
-- Starcoder2(`bigcode/starcoder2-3b`, `bigcode/starcoder2-7b`, `bigcode/starcoder2-15b`, etc.)
-- Xverse (`xverse/XVERSE-7B-Chat`, `xverse/XVERSE-13B-Chat`, `xverse/XVERSE-65B-Chat`, etc.)
-- Yi (`01-ai/Yi-6B`, `01-ai/Yi-34B`, etc.)
 
 Install vLLM with pip or [from source](https://vllm.readthedocs.io/en/latest/getting_started/installation.html#build-from-source):
 
@@ -92,9 +84,7 @@ Install vLLM with pip or [from source](https://vllm.readthedocs.io/en/latest/get
 pip install vllm
 ```
 
-## Getting Started
+Visit our [documentation](https://vllm.readthedocs.io/en/latest/) to learn more.
-
-Visit our [documentation](https://vllm.readthedocs.io/en/latest/) to get started.
 - [Installation](https://vllm.readthedocs.io/en/latest/getting_started/installation.html)
 - [Quickstart](https://vllm.readthedocs.io/en/latest/getting_started/quickstart.html)
 - [Supported Models](https://vllm.readthedocs.io/en/latest/models/supported_models.html)
@@ -104,6 +94,33 @@ Visit our [documentation](https://vllm.readthedocs.io/en/latest/) to get started
 We welcome and value any contributions and collaborations.
 Please check out [CONTRIBUTING.md](./CONTRIBUTING.md) for how to get involved.
 
+## Sponsors
+
+vLLM is a community project. Our compute resources for development and testing are supported by the following organizations. Thank you for your support!
+
+<!-- Note: Please sort them in alphabetical order. -->
+<!-- Note: Please keep these consistent with docs/source/community/sponsors.md -->
+
+- a16z
+- AMD
+- Anyscale
+- AWS
+- Crusoe Cloud
+- Databricks
+- DeepInfra
+- Dropbox
+- Lambda Lab
+- NVIDIA
+- Replicate
+- Roblox
+- RunPod
+- Sequoia Capital
+- Trainy
+- UC Berkeley
+- UC San Diego
+
+We also have an official fundraising venue through [OpenCollective](https://opencollective.com/vllm). We plan to use the fund to support the development, maintenance, and adoption of vLLM.
+
 ## Citation
 
 If you use vLLM for your research, please cite our [paper](https://arxiv.org/abs/2309.06180):

benchmarks/backend_request_func.py

@@ -89,6 +89,9 @@ async def async_request_tgi(
                     output.latency = most_recent_timestamp - st
                     output.success = True
                     output.generated_text = data["generated_text"]
+                else:
+                    output.error = response.reason or ""
+                    output.success = False
         except Exception:
             output.success = False
             exc_info = sys.exc_info()
@@ -276,6 +279,9 @@ async def async_request_openai_completions(
                     output.generated_text = generated_text
                     output.success = True
                     output.latency = latency
+                else:
+                    output.error = response.reason or ""
+                    output.success = False
         except Exception:
             output.success = False
             exc_info = sys.exc_info()

benchmarks/benchmark_latency.py

@@ -1,14 +1,16 @@
 """Benchmark the latency of processing a single batch of requests."""
 import argparse
+import json
 import time
 from pathlib import Path
-from typing import Optional
+from typing import List, Optional
 
 import numpy as np
 import torch
 from tqdm import tqdm
 
 from vllm import LLM, SamplingParams
+from vllm.inputs import PromptStrictInputs
 from vllm.model_executor.layers.quantization import QUANTIZATION_METHODS
 
 
@@ -18,6 +20,8 @@ def main(args: argparse.Namespace):
     # NOTE(woosuk): If the request cannot be processed in a single batch,
     # the engine will automatically process the request in multiple batches.
     llm = LLM(model=args.model,
+              speculative_model=args.speculative_model,
+              num_speculative_tokens=args.num_speculative_tokens,
               tokenizer=args.tokenizer,
               quantization=args.quantization,
               tensor_parallel_size=args.tensor_parallel_size,
@@ -28,9 +32,12 @@ def main(args: argparse.Namespace):
               quantization_param_path=args.quantization_param_path,
               device=args.device,
               ray_workers_use_nsight=args.ray_workers_use_nsight,
+              use_v2_block_manager=args.use_v2_block_manager,
               enable_chunked_prefill=args.enable_chunked_prefill,
               download_dir=args.download_dir,
-              block_size=args.block_size)
+              block_size=args.block_size,
+              gpu_memory_utilization=args.gpu_memory_utilization,
+              distributed_executor_backend=args.distributed_executor_backend)
 
     sampling_params = SamplingParams(
         n=args.n,
@@ -44,7 +51,9 @@ def main(args: argparse.Namespace):
     dummy_prompt_token_ids = np.random.randint(10000,
                                                size=(args.batch_size,
                                                      args.input_len))
-    dummy_prompt_token_ids = dummy_prompt_token_ids.tolist()
+    dummy_inputs: List[PromptStrictInputs] = [{
+        "prompt_token_ids": batch
+    } for batch in dummy_prompt_token_ids.tolist()]
 
     def run_to_completion(profile_dir: Optional[str] = None):
         if profile_dir:
@@ -55,13 +64,13 @@ def main(args: argparse.Namespace):
                     ],
                     on_trace_ready=torch.profiler.tensorboard_trace_handler(
                         str(profile_dir))) as p:
-                llm.generate(prompt_token_ids=dummy_prompt_token_ids,
+                llm.generate(dummy_inputs,
                              sampling_params=sampling_params,
                              use_tqdm=False)
             print(p.key_averages())
         else:
             start_time = time.perf_counter()
-            llm.generate(prompt_token_ids=dummy_prompt_token_ids,
+            llm.generate(dummy_inputs,
                          sampling_params=sampling_params,
                          use_tqdm=False)
             end_time = time.perf_counter()
@@ -93,12 +102,24 @@ def main(args: argparse.Namespace):
     for percentage, percentile in zip(percentages, percentiles):
         print(f'{percentage}% percentile latency: {percentile} seconds')
 
+    # Output JSON results if specified
+    if args.output_json:
+        results = {
+            "avg_latency": np.mean(latencies),
+            "latencies": latencies.tolist(),
+            "percentiles": dict(zip(percentages, percentiles.tolist())),
+        }
+        with open(args.output_json, "w") as f:
+            json.dump(results, f, indent=4)
+
 
 if __name__ == '__main__':
     parser = argparse.ArgumentParser(
         description='Benchmark the latency of processing a single batch of '
         'requests till completion.')
     parser.add_argument('--model', type=str, default='facebook/opt-125m')
+    parser.add_argument('--speculative-model', type=str, default=None)
+    parser.add_argument('--num-speculative-tokens', type=int, default=None)
     parser.add_argument('--tokenizer', type=str, default=None)
     parser.add_argument('--quantization',
                         '-q',
@@ -137,15 +158,13 @@ if __name__ == '__main__':
                         action='store_true',
                         help='enforce eager mode and disable CUDA graph')
     parser.add_argument(
-        "--kv-cache-dtype",
+        '--kv-cache-dtype',
         type=str,
-        choices=['auto', 'fp8'],
+        choices=['auto', 'fp8', 'fp8_e5m2', 'fp8_e4m3'],
-        default='auto',
+        default="auto",
-        help=
+        help='Data type for kv cache storage. If "auto", will use model '
-        'Data type for kv cache storage. If "auto", will use model data type. '
+        'data type. CUDA 11.8+ supports fp8 (=fp8_e4m3) and fp8_e5m2. '
-        'FP8_E5M2 (without scaling) is only supported on cuda version greater '
+        'ROCm (AMD GPU) supports fp8 (=fp8_e4m3)')
-        'than 11.8. On ROCm (AMD GPU), FP8_E4M3 is instead supported for '
-        'common inference criteria.')
     parser.add_argument(
         '--quantization-param-path',
         type=str,
@@ -181,6 +200,7 @@ if __name__ == '__main__':
         action='store_true',
         help='If True, the prefill requests can be chunked based on the '
         'max_num_batched_tokens')
+    parser.add_argument('--use-v2-block-manager', action='store_true')
     parser.add_argument(
         "--ray-workers-use-nsight",
         action='store_true',
@@ -191,5 +211,23 @@ if __name__ == '__main__':
                         default=None,
                         help='directory to download and load the weights, '
                         'default to the default cache dir of huggingface')
+    parser.add_argument(
+        '--output-json',
+        type=str,
+        default=None,
+        help='Path to save the latency results in JSON format.')
+    parser.add_argument('--gpu-memory-utilization',
+                        type=float,
+                        default=0.9,
+                        help='the fraction of GPU memory to be used for '
+                        'the model executor, which can range from 0 to 1.'
+                        'If unspecified, will use the default value of 0.9.')
+    parser.add_argument(
+        '--distributed-executor-backend',
+        choices=['ray', 'mp'],
+        default=None,
+        help='Backend to use for distributed serving. When more than 1 GPU '
+        'is used, will be automatically set to "ray" if installed '
+        'or "mp" (multiprocessing) otherwise.')
     args = parser.parse_args()
     main(args)

benchmarks/benchmark_prefix_caching.py

@@ -16,20 +16,22 @@ def test_prefix(llm=None, sampling_params=None, prompts=None):
 
 
 def main(args):
-    llm = LLM(model="baichuan-inc/Baichuan2-13B-Chat",
+    llm = LLM(model=args.model,
               tokenizer_mode='auto',
               trust_remote_code=True,
               enforce_eager=True,
+              use_v2_block_manager=args.use_v2_block_manager,
+              tensor_parallel_size=args.tensor_parallel_size,
               enable_prefix_caching=args.enable_prefix_caching)
 
     num_prompts = 100
     prompts = [PROMPT] * num_prompts
-    sampling_params = SamplingParams(temperature=0, max_tokens=100)
+    sampling_params = SamplingParams(temperature=0, max_tokens=args.output_len)
 
     print("------warm up------")
     test_prefix(
         llm=llm,
-        prompts=prompts[:1],
+        prompts=prompts,
         sampling_params=sampling_params,
     )
 
@@ -45,8 +47,16 @@ if __name__ == "__main__":
     parser = argparse.ArgumentParser(
         description='Benchmark the performance with or without automatic '
         'prefix caching.')
+    parser.add_argument('--model',
+                        type=str,
+                        default='baichuan-inc/Baichuan2-13B-Chat')
+    parser.add_argument('--tensor-parallel-size', '-tp', type=int, default=1)
+    parser.add_argument('--output-len', type=int, default=10)
     parser.add_argument('--enable-prefix-caching',
                         action='store_true',
                         help='enable prefix caching')
+    parser.add_argument('--use-v2-block-manager',
+                        action='store_true',
+                        help='Use BlockSpaceMangerV2')
     args = parser.parse_args()
     main(args)

benchmarks/benchmark_serving.py

@@ -17,6 +17,10 @@ On the client side, run:
         --dataset-path <path to dataset> \
         --request-rate <request_rate> \ # By default <request_rate> is inf
         --num-prompts <num_prompts> # By default <num_prompts> is 1000
+        
+    when using tgi backend, add
+        --endpoint /generate_stream
+    to the end of the command above.
 """
 import argparse
 import asyncio
@@ -27,7 +31,7 @@ import time
 import warnings
 from dataclasses import dataclass
 from datetime import datetime
-from typing import AsyncGenerator, List, Tuple
+from typing import AsyncGenerator, List, Optional, Tuple
 
 import numpy as np
 from backend_request_func import (ASYNC_REQUEST_FUNCS, RequestFuncInput,
@@ -52,13 +56,20 @@ class BenchmarkMetrics:
     mean_tpot_ms: float
     median_tpot_ms: float
     p99_tpot_ms: float
+    mean_itl_ms: float
+    median_itl_ms: float
+    p99_itl_ms: float
 
 
 def sample_sharegpt_requests(
     dataset_path: str,
     num_requests: int,
     tokenizer: PreTrainedTokenizerBase,
+    fixed_output_len: Optional[int] = None,
 ) -> List[Tuple[str, int, int]]:
+    if fixed_output_len is not None and fixed_output_len < 4:
+        raise ValueError("output_len too small")
+
     # Load the dataset.
     with open(dataset_path) as f:
         dataset = json.load(f)
@@ -68,38 +79,32 @@ def sample_sharegpt_requests(
     dataset = [(data["conversations"][0]["value"],
                 data["conversations"][1]["value"]) for data in dataset]
 
-    # some of these will be filtered out, so sample more than we need
+    # Shuffle the dataset.
-    sampled_indices = random.sample(range(len(dataset)),
+    random.shuffle(dataset)
-                                    int(num_requests * 1.2))
+
-    dataset = [dataset[i] for i in sampled_indices]
+    # Filter out sequences that are too long or too short
+    filtered_dataset: List[Tuple[str, int, int]] = []
+    for i in range(len(dataset)):
+        if len(filtered_dataset) == num_requests:
+            break
 
         # Tokenize the prompts and completions.
-    prompts = [prompt for prompt, _ in dataset]
+        prompt = dataset[i][0]
-    prompt_token_ids = tokenizer(prompts).input_ids
+        prompt_token_ids = tokenizer(prompt).input_ids
-    completions = [completion for _, completion in dataset]
+        completion = dataset[i][1]
-    completion_token_ids = tokenizer(completions).input_ids
+        completion_token_ids = tokenizer(completion).input_ids
-    tokenized_dataset = []
-    for i in range(len(dataset)):
-        output_len = len(completion_token_ids[i])
-        tokenized_dataset.append((prompts[i], prompt_token_ids[i], output_len))
-
-    # Filter out too long sequences.
-    filtered_dataset: List[Tuple[str, int, int]] = []
-    for prompt, prompt_token_ids, output_len in tokenized_dataset:
         prompt_len = len(prompt_token_ids)
+        output_len = len(completion_token_ids
+                         ) if fixed_output_len is None else fixed_output_len
         if prompt_len < 4 or output_len < 4:
             # Prune too short sequences.
-            # This is because TGI causes errors when the input or output length
-            # is too short.
             continue
         if prompt_len > 1024 or prompt_len + output_len > 2048:
             # Prune too long sequences.
             continue
         filtered_dataset.append((prompt, prompt_len, output_len))
 
-    # Sample the requests.
+    return filtered_dataset
-    sampled_requests = random.sample(filtered_dataset, num_requests)
-    return sampled_requests
 
 
 def sample_sonnet_requests(
@@ -198,21 +203,34 @@ def calculate_metrics(
     actual_output_lens = []
     total_input = 0
     completed = 0
+    itls = []
     tpots = []
     ttfts = []
     for i in range(len(outputs)):
         if outputs[i].success:
-            output_len = len(tokenizer(outputs[i].generated_text).input_ids)
+            # We use the tokenizer to count the number of output tokens for all
+            # serving backends instead of looking at len(outputs[i].itl) since
+            # multiple output tokens may be bundled together
+            # Note: this may inflate the output token count slightly
+            output_len = len(
+                tokenizer(outputs[i].generated_text,
+                          add_special_tokens=False).input_ids)
             actual_output_lens.append(output_len)
             total_input += input_requests[i][1]
             if output_len > 1:
                 tpots.append(
                     (outputs[i].latency - outputs[i].ttft) / (output_len - 1))
+            itls += outputs[i].itl
             ttfts.append(outputs[i].ttft)
             completed += 1
         else:
             actual_output_lens.append(0)
 
+    if completed == 0:
+        warnings.warn(
+            "All requests failed. This is likely due to a misconfiguration "
+            "on the benchmark arguments.",
+            stacklevel=2)
     metrics = BenchmarkMetrics(
         completed=completed,
         total_input=total_input,
@@ -224,9 +242,12 @@ def calculate_metrics(
         1000,  # ttfts is empty if streaming is not supported by backend
         median_ttft_ms=np.median(ttfts or 0) * 1000,
         p99_ttft_ms=np.percentile(ttfts or 0, 99) * 1000,
-        mean_tpot_ms=np.mean(tpots) * 1000,
+        mean_tpot_ms=np.mean(tpots or 0) * 1000,
-        median_tpot_ms=np.median(tpots) * 1000,
+        median_tpot_ms=np.median(tpots or 0) * 1000,
-        p99_tpot_ms=np.percentile(tpots, 99) * 1000,
+        p99_tpot_ms=np.percentile(tpots or 0, 99) * 1000,
+        mean_itl_ms=np.mean(itls or 0) * 1000,
+        median_itl_ms=np.median(itls or 0) * 1000,
+        p99_itl_ms=np.percentile(itls or 0, 99) * 1000,
     )
 
     return metrics, actual_output_lens
@@ -248,6 +269,24 @@ async def benchmark(
     else:
         raise ValueError(f"Unknown backend: {backend}")
 
+    print("Starting initial single prompt test run...")
+    test_prompt, test_prompt_len, test_output_len = input_requests[0]
+    test_input = RequestFuncInput(
+        model=model_id,
+        prompt=test_prompt,
+        api_url=api_url,
+        prompt_len=test_prompt_len,
+        output_len=test_output_len,
+        best_of=best_of,
+        use_beam_search=use_beam_search,
+    )
+    test_output = await request_func(request_func_input=test_input)
+    if not test_output.success:
+        raise ValueError(
+            "Initial test run failed - Please make sure benchmark arguments "
+            f"are correctly specified. Error: {test_output.error}")
+    else:
+        print("Initial test run completed. Starting main benchmark run...")
     print(f"Traffic request rate: {request_rate}")
 
     pbar = None if disable_tqdm else tqdm(total=len(input_requests))
@@ -308,6 +347,10 @@ async def benchmark(
     print("{:<40} {:<10.2f}".format("Median TPOT (ms):",
                                     metrics.median_tpot_ms))
     print("{:<40} {:<10.2f}".format("P99 TPOT (ms):", metrics.p99_tpot_ms))
+    print("{s:{c}^{n}}".format(s='Inter-token Latency', n=50, c='-'))
+    print("{:<40} {:<10.2f}".format("Mean ITL (ms):", metrics.mean_itl_ms))
+    print("{:<40} {:<10.2f}".format("Median ITL (ms):", metrics.median_itl_ms))
+    print("{:<40} {:<10.2f}".format("P99 ITL (ms):", metrics.p99_itl_ms))
     print("=" * 50)
 
     result = {
@@ -324,6 +367,9 @@ async def benchmark(
         "mean_tpot_ms": metrics.mean_tpot_ms,
         "median_tpot_ms": metrics.median_tpot_ms,
         "p99_tpot_ms": metrics.p99_tpot_ms,
+        "mean_itl_ms": metrics.mean_itl_ms,
+        "median_itl_ms": metrics.median_itl_ms,
+        "p99_itl_ms": metrics.p99_itl_ms,
         "input_lens": [output.prompt_len for output in outputs],
         "output_lens": actual_output_lens,
         "ttfts": [output.ttft for output in outputs],
@@ -361,6 +407,7 @@ def main(args: argparse.Namespace):
             dataset_path=args.dataset,
             num_requests=args.num_prompts,
             tokenizer=tokenizer,
+            fixed_output_len=args.sharegpt_output_len,
         )
 
     elif args.dataset_name == "sharegpt":
@@ -368,6 +415,7 @@ def main(args: argparse.Namespace):
             dataset_path=args.dataset_path,
             num_requests=args.num_prompts,
             tokenizer=tokenizer,
+            fixed_output_len=args.sharegpt_output_len,
         )
 
     elif args.dataset_name == "sonnet":
@@ -524,6 +572,12 @@ if __name__ == "__main__":
         default=1000,
         help="Number of prompts to process.",
     )
+    parser.add_argument(
+        "--sharegpt-output-len",
+        type=int,
+        default=None,
+        help="Output length for each request. Overrides the output length "
+        "from the ShareGPT dataset.")
     parser.add_argument(
         "--sonnet-input-len",
         type=int,

benchmarks/benchmark_throughput.py

@@ -78,6 +78,7 @@ def run_vllm(
     enable_prefix_caching: bool,
     enable_chunked_prefill: bool,
     max_num_batched_tokens: int,
+    distributed_executor_backend: Optional[str],
     gpu_memory_utilization: float = 0.9,
     download_dir: Optional[str] = None,
 ) -> float:
@@ -100,28 +101,26 @@ def run_vllm(
         download_dir=download_dir,
         enable_chunked_prefill=enable_chunked_prefill,
         max_num_batched_tokens=max_num_batched_tokens,
+        distributed_executor_backend=distributed_executor_backend,
     )
 
     # Add the requests to the engine.
+    prompts = []
+    sampling_params = []
     for prompt, _, output_len in requests:
-        sampling_params = SamplingParams(
+        prompts.append(prompt)
+        sampling_params.append(
+            SamplingParams(
                 n=n,
                 temperature=0.0 if use_beam_search else 1.0,
                 top_p=1.0,
                 use_beam_search=use_beam_search,
                 ignore_eos=True,
                 max_tokens=output_len,
-        )
+            ))
-        # FIXME(woosuk): Do not use internal method.
-        llm._add_request(
-            prompt=prompt,
-            prompt_token_ids=None,
-            sampling_params=sampling_params,
-        )
 
     start = time.perf_counter()
-    # FIXME(woosuk): Do not use internal method.
+    llm.generate(prompts, sampling_params, use_tqdm=True)
-    llm._run_engine(use_tqdm=True)
     end = time.perf_counter()
     return end - start
 
@@ -228,8 +227,8 @@ def main(args: argparse.Namespace):
             args.enforce_eager, args.kv_cache_dtype,
             args.quantization_param_path, args.device,
             args.enable_prefix_caching, args.enable_chunked_prefill,
-            args.max_num_batched_tokens, args.gpu_memory_utilization,
+            args.max_num_batched_tokens, args.distributed_executor_backend,
-            args.download_dir)
+            args.gpu_memory_utilization, args.download_dir)
     elif args.backend == "hf":
         assert args.tensor_parallel_size == 1
         elapsed_time = run_hf(requests, args.model, tokenizer, args.n,
@@ -245,6 +244,18 @@ def main(args: argparse.Namespace):
     print(f"Throughput: {len(requests) / elapsed_time:.2f} requests/s, "
           f"{total_num_tokens / elapsed_time:.2f} tokens/s")
 
+    # Output JSON results if specified
+    if args.output_json:
+        results = {
+            "elapsed_time": elapsed_time,
+            "num_requests": len(requests),
+            "total_num_tokens": total_num_tokens,
+            "requests_per_second": len(requests) / elapsed_time,
+            "tokens_per_second": total_num_tokens / elapsed_time,
+        }
+        with open(args.output_json, "w") as f:
+            json.dump(results, f, indent=4)
+
 
 if __name__ == "__main__":
     parser = argparse.ArgumentParser(description="Benchmark the throughput.")
@@ -314,15 +325,13 @@ if __name__ == "__main__":
                         action="store_true",
                         help="enforce eager execution")
     parser.add_argument(
-        "--kv-cache-dtype",
+        '--kv-cache-dtype',
         type=str,
-        choices=["auto", "fp8"],
+        choices=['auto', 'fp8', 'fp8_e5m2', 'fp8_e4m3'],
         default="auto",
-        help=
+        help='Data type for kv cache storage. If "auto", will use model '
-        'Data type for kv cache storage. If "auto", will use model data type. '
+        'data type. CUDA 11.8+ supports fp8 (=fp8_e4m3) and fp8_e5m2. '
-        'FP8_E5M2 (without scaling) is only supported on cuda version greater '
+        'ROCm (AMD GPU) supports fp8 (=fp8_e4m3)')
-        'than 11.8. On ROCm (AMD GPU), FP8_E4M3 is instead supported for '
-        'common inference criteria.')
     parser.add_argument(
         '--quantization-param-path',
         type=str,
@@ -356,6 +365,18 @@ if __name__ == "__main__":
                         default=None,
                         help='directory to download and load the weights, '
                         'default to the default cache dir of huggingface')
+    parser.add_argument(
+        '--output-json',
+        type=str,
+        default=None,
+        help='Path to save the throughput results in JSON format.')
+    parser.add_argument(
+        '--distributed-executor-backend',
+        choices=['ray', 'mp'],
+        default=None,
+        help='Backend to use for distributed serving. When more than 1 GPU '
+        'is used, will be automatically set to "ray" if installed '
+        'or "mp" (multiprocessing) otherwise.')
     args = parser.parse_args()
     if args.tokenizer is None:
         args.tokenizer = args.model

benchmarks/cutlass_benchmarks/w8a8_benchmarks.py

@@ -0,0 +1,352 @@
+import argparse
+import copy
+import itertools
+import pickle as pkl
+import time
+from typing import Callable, Iterable, List, Tuple
+
+import torch
+import torch.utils.benchmark as TBenchmark
+from torch.utils.benchmark import Measurement as TMeasurement
+from weight_shapes import WEIGHT_SHAPES
+
+from vllm import _custom_ops as ops
+
+DEFAULT_MODELS = list(WEIGHT_SHAPES.keys())[1:]
+DEFAULT_BATCH_SIZES = [1, 16, 32, 64, 128, 256, 512]
+DEFAULT_TP_SIZES = [1]
+
+# helpers
+
+
+def to_fp8(tensor: torch.tensor) -> torch.tensor:
+    finfo = torch.finfo(torch.float8_e4m3fn)
+    return torch.round(tensor.clamp(
+        min=finfo.min, max=finfo.max)).to(dtype=torch.float8_e4m3fn)
+
+
+def to_int8(tensor: torch.tensor) -> torch.tensor:
+    return torch.round(tensor.clamp(min=-128, max=127)).to(dtype=torch.int8)
+
+
+def make_rand_tensors(dtype: torch.dtype, m: int, n: int,
+                      k: int) -> Tuple[torch.tensor, torch.tensor]:
+
+    a = torch.randn((m, k), device='cuda') * 5
+    b = torch.randn((n, k), device='cuda').t() * 5
+
+    if dtype == torch.int8:
+        return to_int8(a), to_int8(b)
+    if dtype == torch.float8_e4m3fn:
+        return to_fp8(a), to_fp8(b)
+
+    raise ValueError("unsupported dtype")
+
+
+# impl
+
+
+def pytorch_i8_impl(a: torch.tensor, b: torch.tensor, scale_a: torch.tensor,
+                    scale_b: torch.tensor,
+                    out_dtype: torch.dtype) -> torch.tensor:
+    return torch.mm(a, b)
+
+
+def pytorch_fp8_impl(a: torch.tensor, b: torch.tensor, scale_a: torch.tensor,
+                     scale_b: torch.tensor,
+                     out_dtype: torch.dtype) -> torch.tensor:
+    return torch._scaled_mm(a,
+                            b,
+                            scale_a=scale_a,
+                            scale_b=scale_b,
+                            out_dtype=out_dtype)
+
+
+def pytorch_fp8_impl_fast_accum(a: torch.tensor, b: torch.tensor,
+                                scale_a: torch.tensor, scale_b: torch.tensor,
+                                out_dtype: torch.dtype) -> torch.tensor:
+    return torch._scaled_mm(a,
+                            b,
+                            scale_a=scale_a,
+                            scale_b=scale_b,
+                            out_dtype=out_dtype,
+                            use_fast_accum=True)
+
+
+def cutlass_impl(a: torch.tensor, b: torch.tensor, scale_a: torch.tensor,
+                 scale_b: torch.tensor,
+                 out_dtype: torch.dtype) -> torch.tensor:
+    return ops.cutlass_scaled_mm_dq(a,
+                                    b,
+                                    scale_a,
+                                    scale_b,
+                                    out_dtype=out_dtype)
+
+
+# bench
+def bench_fn(a: torch.tensor, b: torch.tensor, scale_a: torch.tensor,
+             scale_b: torch.tensor, out_dtype: torch.dtype, label: str,
+             sub_label: str, fn: Callable, description: str) -> TMeasurement:
+
+    min_run_time = 1
+
+    globals = {
+        "a": a,
+        "b": b,
+        "scale_a": scale_a,
+        "scale_b": scale_b,
+        "out_dtype": out_dtype,
+        "fn": fn,
+    }
+    return TBenchmark.Timer(
+        stmt="fn(a, b, scale_a, scale_b, out_dtype)",
+        globals=globals,
+        label=label,
+        sub_label=sub_label,
+        description=description,
+    ).blocked_autorange(min_run_time=min_run_time)
+
+
+def bench_int8(dtype: torch.dtype, m: int, k: int, n: int, label: str,
+               sub_label: str) -> Iterable[TMeasurement]:
+    assert dtype == torch.int8
+    a, b = make_rand_tensors(torch.int8, m, n, k)
+    scale_a = torch.tensor(1.0, device="cuda", dtype=torch.float32)
+    scale_b = torch.tensor(1.0, device="cuda", dtype=torch.float32)
+
+    timers = []
+    # pytorch impl
+    timers.append(
+        bench_fn(a.to(dtype=torch.bfloat16, device="cuda"),
+                 b.to(dtype=torch.bfloat16, device="cuda"), scale_a, scale_b,
+                 torch.bfloat16, label, sub_label, pytorch_i8_impl,
+                 "pytorch_bf16_bf16_bf16_matmul-no-scales"))
+
+    # cutlass impl
+    timers.append(
+        bench_fn(a, b, scale_a.to(device="cpu"), scale_b.to(device="cpu"),
+                 torch.bfloat16, label, sub_label, cutlass_impl,
+                 "cutlass_i8_i8_bf16_scaled_mm"))
+
+    return timers
+
+
+def bench_fp8(dtype: torch.dtype, m: int, k: int, n: int, label: str,
+              sub_label: str) -> Iterable[TMeasurement]:
+    assert dtype == torch.float8_e4m3fn
+    a, b = make_rand_tensors(torch.float8_e4m3fn, m, n, k)
+    scale_a = torch.tensor(1.0, device="cuda", dtype=torch.float32)
+    scale_b = torch.tensor(1.0, device="cuda", dtype=torch.float32)
+
+    timers = []
+
+    # pytorch impl: bf16 output, without fp8 fast accum
+    timers.append(
+        bench_fn(a, b, scale_a, scale_b, torch.bfloat16, label, sub_label,
+                 pytorch_fp8_impl, "pytorch_fp8_fp8_bf16_scaled_mm"))
+
+    # pytorch impl: bf16 output, with fp8 fast accum
+    timers.append(
+        bench_fn(a, b, scale_a, scale_b, torch.bfloat16, label, sub_label,
+                 pytorch_fp8_impl_fast_accum,
+                 "pytorch_fp8_fp8_bf16_scaled_mm_fast_accum"))
+
+    # pytorch impl: fp16 output, without fp8 fast accum
+    timers.append(
+        bench_fn(a, b, scale_a, scale_b, torch.float16, label, sub_label,
+                 pytorch_fp8_impl, "pytorch_fp8_fp8_fp16_scaled_mm"))
+
+    # pytorch impl: fp16 output, with fp8 fast accum
+    timers.append(
+        bench_fn(a, b, scale_a, scale_b, torch.float16, label, sub_label,
+                 pytorch_fp8_impl_fast_accum,
+                 "pytorch_fp8_fp8_fp16_scaled_mm_fast_accum"))
+
+    # cutlass impl: bf16 output
+    timers.append(
+        bench_fn(a, b, scale_a.to(device="cpu"), scale_b.to(device="cpu"),
+                 torch.bfloat16, label, sub_label, cutlass_impl,
+                 "cutlass_fp8_fp8_bf16_scaled_mm"))
+    # cutlass impl: fp16 output
+    timers.append(
+        bench_fn(a, b, scale_a.to(device="cpu"), scale_b.to(device="cpu"),
+                 torch.float16, label, sub_label, cutlass_impl,
+                 "cutlass_fp8_fp8_fp16_scaled_mm"))
+    return timers
+
+
+def bench(dtype: torch.dtype, m: int, k: int, n: int, label: str,
+          sub_label: str) -> Iterable[TMeasurement]:
+    if dtype == torch.int8:
+        return bench_int8(dtype, m, k, n, label, sub_label)
+    if dtype == torch.float8_e4m3fn:
+        return bench_fp8(dtype, m, k, n, label, sub_label)
+    raise ValueError("unsupported type")
+
+
+# runner
+def print_timers(timers: Iterable[TMeasurement]):
+    compare = TBenchmark.Compare(timers)
+    compare.print()
+
+
+def run(dtype: torch.dtype,
+        MKNs: Iterable[Tuple[int, int, int]]) -> Iterable[TMeasurement]:
+
+    results = []
+    for m, k, n in MKNs:
+        timers = bench(dtype, m, k, n, f"scaled-{dtype}-gemm",
+                       f"MKN=({m}x{k}x{n})")
+        print_timers(timers)
+        results.extend(timers)
+
+    return results
+
+
+# output makers
+def make_output(data: Iterable[TMeasurement],
+                MKNs: Iterable[Tuple[int, int, int]],
+                base_description: str,
+                timestamp=None):
+
+    print(f"== All Results {base_description} ====")
+    print_timers(data)
+
+    # pickle all the results
+    timestamp = int(time.time()) if timestamp is None else timestamp
+    with open(f"{base_description}-{timestamp}.pkl", "wb") as f:
+        pkl.dump(data, f)
+
+
+# argparse runners
+
+
+def run_square_bench(args):
+    dim_sizes = list(
+        range(args.dim_start, args.dim_end + 1, args.dim_increment))
+    MKNs = list(zip(dim_sizes, dim_sizes, dim_sizes))
+    data = run(args.dtype, MKNs)
+
+    make_output(data, MKNs, f"square_bench-{args.dtype}")
+
+
+def run_range_bench(args):
+    dim_sizes = list(range(args.dim_start, args.dim_end, args.dim_increment))
+    n = len(dim_sizes)
+    Ms = [args.m_constant] * n if args.m_constant is not None else dim_sizes
+    Ks = [args.k_constant] * n if args.k_constant is not None else dim_sizes
+    Ns = [args.n_constant] * n if args.n_constant is not None else dim_sizes
+    MKNs = list(zip(Ms, Ks, Ns))
+    data = run(args.dtype, MKNs)
+
+    make_output(data, MKNs, f"range_bench-{args.dtype}")
+
+
+def run_model_bench(args):
+
+    print("Benchmarking models:")
+    for i, model in enumerate(args.models):
+        print(f"[{i}]  {model}")
+
+    def model_shapes(model_name: str, tp_size: int) -> List[Tuple[int, int]]:
+        KNs = []
+        for KN, tp_split_dim in copy.deepcopy(WEIGHT_SHAPES[model_name]):
+            KN[tp_split_dim] = KN[tp_split_dim] // tp_size
+            KNs.append(KN)
+        return KNs
+
+    model_bench_data = []
+    models_tps = list(itertools.product(args.models, args.tp_sizes))
+    for model, tp_size in models_tps:
+        Ms = args.batch_sizes
+        KNs = model_shapes(model, tp_size)
+        MKNs = []
+        for m in Ms:
+            for k, n in KNs:
+                MKNs.append((m, k, n))
+
+        data = run(args.dtype, MKNs)
+        model_bench_data.append(data)
+
+    # Print all results
+    for data, model_tp in zip(model_bench_data, models_tps):
+        model, tp_size = model_tp
+        print(f"== Results {args.dtype} {model}-TP{tp_size} ====")
+        print_timers(data)
+
+    timestamp = int(time.time())
+
+    all_data = []
+    for d in model_bench_data:
+        all_data.extend(d)
+    # pickle all data
+    with open(f"model_bench-{args.dtype}-{timestamp}.pkl", "wb") as f:
+        pkl.dump(all_data, f)
+
+
+if __name__ == '__main__':
+
+    def to_torch_dtype(dt):
+        if dt == "int8":
+            return torch.int8
+        if dt == "fp8":
+            return torch.float8_e4m3fn
+        raise ValueError("unsupported dtype")
+
+    parser = argparse.ArgumentParser(
+        description="""
+Benchmark Cutlass GEMM.
+
+    To run square GEMMs:
+        python3 ./benchmarks/cutlass_benchmarks/w8a8_benchmarks.py --dtype fp8 square_bench --dim-start 128 --dim-end 512 --dim-increment 64
+    
+    To run constant N and K and sweep M:
+        python3 ./benchmarks/cutlass_benchmarks/w8a8_benchmarks.py --dtype fp8 range_bench --dim-start 128 --dim-end 512 --dim-increment 64 --n-constant 16384 --k-constant 16384
+    
+    To run dimensions from a model:
+        python3 ./benchmarks/cutlass_benchmarks/w8a8_benchmarks.py --dtype fp8 model_bench --models meta-llama/Llama-2-7b-hf --batch-sizes 16 --tp-sizes 1
+    
+    Output:
+        - a .pkl file, that is a list of raw torch.benchmark.utils.Measurements for the pytorch and cutlass implementations for the various GEMMs.
+            """,  # noqa: E501
+        formatter_class=argparse.RawTextHelpFormatter)
+
+    parser.add_argument("--dtype",
+                        type=to_torch_dtype,
+                        required=True,
+                        help="Available options are ['int8', 'fp8']")
+    subparsers = parser.add_subparsers(dest="cmd")
+
+    square_parser = subparsers.add_parser("square_bench")
+    square_parser.add_argument("--dim-start", type=int, required=True)
+    square_parser.add_argument("--dim-end", type=int, required=True)
+    square_parser.add_argument("--dim-increment", type=int, required=True)
+    square_parser.set_defaults(func=run_square_bench)
+
+    range_parser = subparsers.add_parser("range_bench")
+    range_parser.add_argument("--dim-start", type=int, required=True)
+    range_parser.add_argument("--dim-end", type=int, required=True)
+    range_parser.add_argument("--dim-increment", type=int, required=True)
+    range_parser.add_argument("--m-constant", type=int, default=None)
+    range_parser.add_argument("--n-constant", type=int, default=None)
+    range_parser.add_argument("--k-constant", type=int, default=None)
+    range_parser.set_defaults(func=run_range_bench)
+
+    model_parser = subparsers.add_parser("model_bench")
+    model_parser.add_argument("--models",
+                              nargs="+",
+                              type=str,
+                              default=DEFAULT_MODELS,
+                              choices=WEIGHT_SHAPES.keys())
+    model_parser.add_argument("--tp-sizes",
+                              nargs="+",
+                              type=int,
+                              default=DEFAULT_TP_SIZES)
+    model_parser.add_argument("--batch-sizes",
+                              nargs="+",
+                              type=int,
+                              default=DEFAULT_BATCH_SIZES)
+    model_parser.set_defaults(func=run_model_bench)
+
+    args = parser.parse_args()
+    args.func(args)

benchmarks/cutlass_benchmarks/weight_shapes.py

@@ -0,0 +1,37 @@
+# Weight Shapes are in the format
+# ([K, N], TP_SPLIT_DIM)
+# Example:
+#  A shape of ([14336, 4096], 0) indicates the following GEMM shape,
+#   - TP1 : K = 14336, N = 4096
+#   - TP2 : K = 7168, N = 4096
+#  A shape of ([4096, 6144], 1) indicates the following GEMM shape,
+#   - TP1 : K = 4096, N = 6144
+#   - TP4 : K = 4096, N = 1536
+
+# TP1 shapes
+WEIGHT_SHAPES = {
+    "mistralai/Mistral-7B-v0.1": [
+        ([4096, 6144], 1),
+        ([4096, 4096], 0),
+        ([4096, 28672], 1),
+        ([14336, 4096], 0),
+    ],
+    "meta-llama/Llama-2-7b-hf": [
+        ([4096, 12288], 1),
+        ([4096, 4096], 0),
+        ([4096, 22016], 1),
+        ([11008, 4096], 0),
+    ],
+    "meta-llama/Llama-2-13b-hf": [
+        ([5120, 15360], 1),
+        ([5120, 5120], 0),
+        ([5120, 27648], 1),
+        ([13824, 5120], 0),
+    ],
+    "meta-llama/Llama-2-70b-hf": [
+        ([8192, 10240], 1),
+        ([8192, 8192], 0),
+        ([8192, 57344], 1),
+        ([28672, 8192], 0),
+    ],
+}

benchmarks/kernels/benchmark_aqlm.py

@@ -6,7 +6,7 @@ from typing import Optional
 import torch
 import torch.nn.functional as F
 
-from vllm._C import ops
+from vllm import _custom_ops as ops
 from vllm.model_executor.layers.quantization.aqlm import (
     dequantize_weight, generic_dequantize_gemm, get_int_dtype,
     optimized_dequantize_gemm)

benchmarks/kernels/benchmark_marlin.py

@@ -0,0 +1,233 @@
+import argparse
+
+import torch
+import torch.utils.benchmark as benchmark
+from benchmark_shapes import WEIGHT_SHAPES
+
+from vllm import _custom_ops as ops
+from vllm.model_executor.layers.quantization.gptq_marlin import (
+    GPTQ_MARLIN_MAX_PARALLEL, GPTQ_MARLIN_MIN_THREAD_N,
+    GPTQ_MARLIN_SUPPORTED_GROUP_SIZES, GPTQ_MARLIN_SUPPORTED_NUM_BITS)
+from vllm.model_executor.layers.quantization.gptq_marlin_24 import (
+    GPTQ_MARLIN_24_MAX_PARALLEL, GPTQ_MARLIN_24_MIN_THREAD_N,
+    GPTQ_MARLIN_24_SUPPORTED_GROUP_SIZES, GPTQ_MARLIN_24_SUPPORTED_NUM_BITS)
+from vllm.model_executor.layers.quantization.utils.marlin_utils import (
+    MarlinWorkspace, marlin_24_quantize, marlin_quantize)
+from vllm.model_executor.layers.quantization.utils.quant_utils import (
+    gptq_pack, quantize_weights, sort_weights)
+
+DEFAULT_MODELS = ["meta-llama/Llama-2-7b-hf/TP1"]
+DEFAULT_BATCH_SIZES = [1, 16, 32, 64, 128, 256, 512]
+
+ACT_ORDER_OPTS = [False, True]
+K_FULL_OPTS = [False, True]
+
+
+def bench_run(results, model, act_order, is_k_full, num_bits, group_size,
+              size_m, size_k, size_n):
+    label = "Quant Matmul"
+
+    sub_label = ("{}, act={} k_full={}, b={}, g={}, "
+                 "MKN=({}x{}x{})".format(model, act_order, is_k_full, num_bits,
+                                         group_size, size_m, size_k, size_n))
+
+    print(f"Testing: {sub_label}")
+
+    a = torch.randn(size_m, size_k).to(torch.half).cuda()
+    b = torch.rand(size_k, size_n).to(torch.half).cuda()
+
+    a_tmp = (torch.zeros(size_m, size_k).to(torch.half).cuda())
+
+    # Marlin quant
+    (
+        marlin_w_ref,
+        marlin_q_w,
+        marlin_s,
+        marlin_g_idx,
+        marlin_sort_indices,
+        marlin_rand_perm,
+    ) = marlin_quantize(b, num_bits, group_size, act_order)
+
+    # Marlin_24 quant
+    (marlin_24_w_ref, marlin_24_q_w_comp, marlin_24_meta,
+     marlin_24_s) = marlin_24_quantize(b, num_bits, group_size)
+
+    # GPTQ quant
+    (w_ref, q_w, s, g_idx,
+     rand_perm) = quantize_weights(b, num_bits, group_size, act_order)
+    q_w_gptq = gptq_pack(q_w, num_bits, size_k, size_n)
+
+    # For act_order, sort the "weights" and "g_idx"
+    # so that group ids are increasing
+    repack_sort_indices = torch.empty(0, dtype=torch.int, device=b.device)
+    if act_order:
+        (q_w, g_idx, repack_sort_indices) = sort_weights(q_w, g_idx)
+
+    # Prepare
+    marlin_workspace = MarlinWorkspace(size_n, GPTQ_MARLIN_MIN_THREAD_N,
+                                       GPTQ_MARLIN_MAX_PARALLEL)
+
+    marlin_24_workspace = MarlinWorkspace(size_n, GPTQ_MARLIN_24_MIN_THREAD_N,
+                                          GPTQ_MARLIN_24_MAX_PARALLEL)
+
+    globals = {
+        # Gen params
+        "num_bits": num_bits,
+        "group_size": group_size,
+        "size_m": size_m,
+        "size_n": size_n,
+        "size_k": size_k,
+        "a": a,
+        "a_tmp": a_tmp,
+        # Marlin params
+        "marlin_w_ref": marlin_w_ref,
+        "marlin_q_w": marlin_q_w,
+        "marlin_s": marlin_s,
+        "marlin_g_idx": marlin_g_idx,
+        "marlin_sort_indices": marlin_sort_indices,
+        "marlin_rand_perm": marlin_rand_perm,
+        "marlin_workspace": marlin_workspace,
+        "is_k_full": is_k_full,
+        # Marlin_24 params
+        "marlin_24_w_ref": marlin_24_w_ref,
+        "marlin_24_q_w_comp": marlin_24_q_w_comp,
+        "marlin_24_meta": marlin_24_meta,
+        "marlin_24_s": marlin_24_s,
+        "marlin_24_workspace": marlin_24_workspace,
+        # GPTQ params
+        "q_w_gptq": q_w_gptq,
+        "repack_sort_indices": repack_sort_indices,
+        # Kernels
+        "gptq_marlin_gemm": ops.gptq_marlin_gemm,
+        "gptq_marlin_24_gemm": ops.gptq_marlin_24_gemm,
+        "gptq_marlin_repack": ops.gptq_marlin_repack,
+    }
+
+    min_run_time = 1
+
+    # Warmup pytorch
+    for i in range(5):
+        torch.matmul(a, marlin_w_ref)
+
+    results.append(
+        benchmark.Timer(
+            stmt="torch.matmul(a, marlin_w_ref)",
+            globals=globals,
+            label=label,
+            sub_label=sub_label,
+            description="pytorch_gemm",
+        ).blocked_autorange(min_run_time=min_run_time))
+
+    results.append(
+        benchmark.Timer(
+            stmt=
+            "output = gptq_marlin_gemm(a, marlin_q_w, marlin_s, marlin_g_idx, marlin_sort_indices, marlin_workspace.scratch, num_bits, size_m, size_n, size_k, is_k_full)",  # noqa: E501
+            globals=globals,
+            label=label,
+            sub_label=sub_label,
+            description="gptq_marlin_gemm",
+        ).blocked_autorange(min_run_time=min_run_time))
+
+    if (num_bits in GPTQ_MARLIN_24_SUPPORTED_NUM_BITS
+            and group_size in GPTQ_MARLIN_24_SUPPORTED_GROUP_SIZES):
+        results.append(
+            benchmark.Timer(
+                stmt=
+                "output = gptq_marlin_24_gemm(a, marlin_24_q_w_comp, marlin_24_meta, marlin_24_s, marlin_24_workspace.scratch, num_bits, size_m, size_n, size_k)",  # noqa: E501
+                globals=globals,
+                label=label,
+                sub_label=sub_label,
+                description="gptq_marlin_24_gemm",
+            ).blocked_autorange(min_run_time=min_run_time))
+
+    results.append(
+        benchmark.Timer(
+            stmt=
+            "q_res = gptq_marlin_repack(q_w_gptq, repack_sort_indices, size_k, size_n, num_bits)",  # noqa: E501
+            globals=globals,
+            label=label,
+            sub_label=sub_label,
+            description="gptq_marlin_repack",
+        ).blocked_autorange(min_run_time=min_run_time))
+
+
+def main(args):
+    print("Benchmarking models:")
+    for i, model in enumerate(args.models):
+        print(f"[{i}]  {model}")
+
+    results = []
+
+    for model in args.models:
+        for layer in WEIGHT_SHAPES[model]:
+            size_k = layer[0]
+            size_n = layer[1]
+
+            if len(args.limit_k) > 0 and size_k not in args.limit_k:
+                continue
+
+            if len(args.limit_n) > 0 and size_n not in args.limit_n:
+                continue
+
+            for act_order in ACT_ORDER_OPTS:
+                if len(args.limit_act_order
+                       ) > 0 and act_order not in args.limit_act_order:
+                    continue
+
+                for is_k_full in K_FULL_OPTS:
+                    if len(args.limit_k_full
+                           ) > 0 and is_k_full not in args.limit_k_full:
+                        continue
+
+                    for num_bits in GPTQ_MARLIN_SUPPORTED_NUM_BITS:
+                        if len(args.limit_num_bits
+                               ) > 0 and num_bits not in args.limit_num_bits:
+                            continue
+
+                        for group_size in GPTQ_MARLIN_SUPPORTED_GROUP_SIZES:
+                            if len(
+                                    args.limit_group_size
+                            ) > 0 and group_size not in args.limit_group_size:
+                                continue
+
+                            # For act_order, the group_size must be less than
+                            # size_k
+                            if act_order and (group_size == size_k
+                                              or group_size == -1):
+                                continue
+
+                            for size_m in args.batch_sizes:
+                                bench_run(results, model, act_order, is_k_full,
+                                          num_bits, group_size, size_m, size_k,
+                                          size_n)
+
+    compare = benchmark.Compare(results)
+    compare.print()
+
+
+# For quick benchmarking use:
+#   python benchmark_marlin.py --batch-sizes 1 16 32 --limit-k 4096 --limit-n 4096 --limit-group-size 128 --limit-num-bits 4 --limit-act-order 0 --limit-k-full 1 # noqa E501
+#
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description="Benchmark Marlin across specified models/shapes/batches")
+    parser.add_argument(
+        "--models",
+        nargs="+",
+        type=str,
+        default=DEFAULT_MODELS,
+        choices=WEIGHT_SHAPES.keys(),
+    )
+    parser.add_argument("--batch-sizes",
+                        nargs="+",
+                        type=int,
+                        default=DEFAULT_BATCH_SIZES)
+    parser.add_argument("--limit-k", nargs="+", type=int, default=[])
+    parser.add_argument("--limit-n", nargs="+", type=int, default=[])
+    parser.add_argument("--limit-group-size", nargs="+", type=int, default=[])
+    parser.add_argument("--limit-num-bits", nargs="+", type=int, default=[])
+    parser.add_argument("--limit-act-order", nargs="+", type=int, default=[])
+    parser.add_argument("--limit-k-full", nargs="+", type=int, default=[])
+
+    args = parser.parse_args()
+    main(args)

benchmarks/kernels/benchmark_mixtral_moe.py

@@ -1,182 +0,0 @@
-import json
-import os
-import sys
-
-import torch
-import torch.nn.functional as F
-import triton
-
-from vllm.model_executor.layers.fused_moe import (fused_moe,
-                                                  get_config_file_name)
-
-os.environ['CUDA_VISIBLE_DEVICES'] = '0'
-
-
-def main():
-    method = fused_moe
-    for bs in [
-            1, 2, 4, 8, 16, 24, 32, 48, 64, 96, 128, 256, 512, 1024, 1536,
-            2048, 3072, 4096
-    ]:
-        run_grid(bs, method=method)
-
-
-def run_grid(bs, method):
-    d_model = 4096
-    num_total_experts = 8
-    top_k = 2
-    tp_size = 2
-    model_intermediate_size = 14336
-    num_layers = 32
-    num_calls = 100
-
-    num_warmup_trials = 1
-    num_trials = 1
-
-    configs = []
-    if bs <= 16:
-        BLOCK_SIZES_M = [16]
-    elif bs <= 32:
-        BLOCK_SIZES_M = [16, 32]
-    elif bs <= 64:
-        BLOCK_SIZES_M = [16, 32, 64]
-    elif bs <= 128:
-        BLOCK_SIZES_M = [16, 32, 64, 128]
-    else:
-        BLOCK_SIZES_M = [16, 32, 64, 128, 256]
-
-    for block_size_n in [32, 64, 128, 256]:
-        for block_size_m in BLOCK_SIZES_M:
-            for block_size_k in [64, 128, 256]:
-                for group_size_m in [1, 16, 32, 64]:
-                    for num_warps in [4, 8]:
-                        configs.append({
-                            "BLOCK_SIZE_M": block_size_m,
-                            "BLOCK_SIZE_N": block_size_n,
-                            "BLOCK_SIZE_K": block_size_k,
-                            "GROUP_SIZE_M": group_size_m,
-                            "num_warps": num_warps,
-                            "num_stages": 4,
-                        })
-
-    best_config = None
-    best_time_us = 1e20
-
-    for config in configs:
-        print(f'{tp_size=} {bs=}')
-        print(f'{config}')
-        # warmup
-        print('warming up')
-        try:
-            for _ in range(num_warmup_trials):
-                run_timing(
-                    num_calls=num_calls,
-                    bs=bs,
-                    d_model=d_model,
-                    num_total_experts=num_total_experts,
-                    top_k=top_k,
-                    tp_size=tp_size,
-                    model_intermediate_size=model_intermediate_size,
-                    method=method,
-                    config=config,
-                )
-        except triton.runtime.autotuner.OutOfResources:
-            continue
-
-        # trial
-        print('benchmarking')
-        for _ in range(num_trials):
-            kernel_dur_ms = run_timing(
-                num_calls=num_calls,
-                bs=bs,
-                d_model=d_model,
-                num_total_experts=num_total_experts,
-                top_k=top_k,
-                tp_size=tp_size,
-                model_intermediate_size=model_intermediate_size,
-                method=method,
-                config=config,
-            )
-
-            kernel_dur_us = 1000 * kernel_dur_ms
-            model_dur_ms = kernel_dur_ms * num_layers
-
-            if kernel_dur_us < best_time_us:
-                best_config = config
-                best_time_us = kernel_dur_us
-
-            print(f'{kernel_dur_us=:.1f} {model_dur_ms=:.1f}'
-                  f' {bs=} {tp_size=} {top_k=} {num_total_experts=} '
-                  f'{d_model=} {model_intermediate_size=} {num_layers=}')
-
-    print("best_time_us", best_time_us)
-    print("best_config", best_config)
-
-    # holds Dict[str, Dict[str, int]]
-    filename = get_config_file_name(num_total_experts,
-                                    model_intermediate_size // tp_size)
-    print(f"writing config to file {filename}")
-    existing_content = {}
-    if os.path.exists(filename):
-        with open(filename, "r") as f:
-            existing_content = json.load(f)
-    existing_content[str(bs)] = best_config
-    with open(filename, "w") as f:
-        json.dump(existing_content, f, indent=4)
-        f.write("\n")
-
-
-def run_timing(num_calls: int, bs: int, d_model: int, num_total_experts: int,
-               top_k: int, tp_size: int, model_intermediate_size: int, method,
-               config) -> float:
-    shard_intermediate_size = model_intermediate_size // tp_size
-
-    hidden_states = torch.rand(
-        (bs, d_model),
-        device="cuda:0",
-        dtype=torch.bfloat16,
-    )
-
-    ws = torch.rand(
-        (num_total_experts, 2 * shard_intermediate_size, d_model),
-        device=hidden_states.device,
-        dtype=hidden_states.dtype,
-    )
-
-    w2s = torch.rand(
-        (num_total_experts, d_model, shard_intermediate_size),
-        device=hidden_states.device,
-        dtype=hidden_states.dtype,
-    )
-
-    gating_output = F.softmax(torch.rand(
-        (num_calls, bs, num_total_experts),
-        device=hidden_states.device,
-        dtype=torch.float32,
-    ),
-                              dim=-1)
-
-    start_event = torch.cuda.Event(enable_timing=True)
-    end_event = torch.cuda.Event(enable_timing=True)
-
-    start_event.record()
-    for i in range(num_calls):
-        hidden_states = method(
-            hidden_states=hidden_states,
-            w1=ws,
-            w2=w2s,
-            gating_output=gating_output[i],
-            topk=2,
-            renormalize=True,
-            inplace=True,
-            override_config=config,
-        )
-    end_event.record()
-    end_event.synchronize()
-
-    dur_ms = start_event.elapsed_time(end_event) / num_calls
-    return dur_ms
-
-
-if __name__ == "__main__":
-    sys.exit(main())

benchmarks/kernels/benchmark_moe.py

@@ -0,0 +1,322 @@
+import argparse
+import time
+from datetime import datetime
+from typing import Any, Dict, List, Tuple
+
+import ray
+import torch
+import triton
+from ray.experimental.tqdm_ray import tqdm
+from transformers import AutoConfig
+
+from vllm.model_executor.layers.fused_moe.fused_moe import *
+
+
+def benchmark_config(
+    config: Dict[str, int],
+    num_tokens: int,
+    num_experts: int,
+    shard_intermediate_size: int,
+    hidden_size: int,
+    topk: int,
+    dtype: torch.dtype,
+    use_fp8: bool,
+    num_iters: int = 100,
+) -> float:
+    init_dtype = torch.float16 if use_fp8 else dtype
+    x = torch.randn(num_tokens, hidden_size, dtype=dtype)
+    w1 = torch.randn(num_experts,
+                     shard_intermediate_size,
+                     hidden_size,
+                     dtype=init_dtype)
+    w2 = torch.randn(num_experts,
+                     hidden_size,
+                     shard_intermediate_size // 2,
+                     dtype=init_dtype)
+    gating_output = torch.randn(num_iters,
+                                num_tokens,
+                                num_experts,
+                                dtype=torch.float32)
+
+    w1_scale = None
+    w2_scale = None
+    a1_scale = None
+    a2_scale = None
+    if use_fp8:
+        w1_scale = torch.randn(num_experts, dtype=torch.float32)
+        w2_scale = torch.randn(num_experts, dtype=torch.float32)
+        a1_scale = torch.randn(1, dtype=torch.float32)
+        a2_scale = torch.randn(1, dtype=torch.float32)
+
+        w1 = w1.to(torch.float8_e4m3fn)
+        w2 = w2.to(torch.float8_e4m3fn)
+
+    input_gating = torch.empty(num_tokens, num_experts, dtype=torch.float32)
+
+    def prepare(i: int):
+        input_gating.copy_(gating_output[i])
+
+    def run():
+        fused_moe(
+            x,
+            w1,
+            w2,
+            input_gating,
+            topk,
+            renormalize=True,
+            inplace=True,
+            override_config=config,
+            use_fp8=use_fp8,
+            w1_scale=w1_scale,
+            w2_scale=w2_scale,
+            a1_scale=a1_scale,
+            a2_scale=a2_scale,
+        )
+
+    # JIT compilation & warmup
+    run()
+    torch.cuda.synchronize()
+
+    # Capture 10 invocations with CUDA graph
+    graph = torch.cuda.CUDAGraph()
+    with torch.cuda.graph(graph):
+        for _ in range(10):
+            run()
+    torch.cuda.synchronize()
+
+    # Warmup
+    for _ in range(5):
+        graph.replay()
+    torch.cuda.synchronize()
+
+    start_event = torch.cuda.Event(enable_timing=True)
+    end_event = torch.cuda.Event(enable_timing=True)
+
+    latencies = []
+    for i in range(num_iters):
+        prepare(i)
+        torch.cuda.synchronize()
+
+        start_event.record()
+        graph.replay()
+        end_event.record()
+        end_event.synchronize()
+        latencies.append(start_event.elapsed_time(end_event))
+    avg = sum(latencies) / (num_iters * 10) * 1000  # us
+    graph.reset()
+    return avg
+
+
+def get_configs_compute_bound() -> List[Dict[str, int]]:
+    # Reduced search space for faster tuning.
+    # TODO(woosuk): Increase the search space and use a performance model to
+    # prune the search space.
+    configs = []
+    for num_stages in [2, 3, 4, 5]:
+        for block_m in [16, 32, 64, 128, 256]:
+            for block_k in [64, 128, 256]:
+                for block_n in [32, 64, 128, 256]:
+                    for num_warps in [4, 8]:
+                        for group_size in [1, 16, 32, 64]:
+                            configs.append({
+                                "BLOCK_SIZE_M": block_m,
+                                "BLOCK_SIZE_N": block_n,
+                                "BLOCK_SIZE_K": block_k,
+                                "GROUP_SIZE_M": group_size,
+                                "num_warps": num_warps,
+                                "num_stages": num_stages,
+                            })
+    return configs
+
+
+@ray.remote(num_gpus=1)
+class BenchmarkWorker:
+
+    def __init__(self, seed: int) -> None:
+        torch.set_default_device("cuda")
+        torch.cuda.manual_seed_all(seed)
+        self.seed = seed
+
+    def benchmark(
+        self,
+        num_tokens: int,
+        num_experts: int,
+        shard_intermediate_size: int,
+        hidden_size: int,
+        topk: int,
+        dtype: torch.dtype,
+        use_fp8: bool,
+    ) -> Tuple[Dict[str, int], float]:
+        torch.cuda.manual_seed_all(self.seed)
+
+        dtype_str = "float8" if use_fp8 else None
+        # NOTE(woosuk): The current naming convention uses w2.shape[2], which
+        # is the intermediate size after silu_and_mul.
+        op_config = get_moe_configs(num_experts, shard_intermediate_size // 2,
+                                    dtype_str)
+        if op_config is None:
+            config = get_default_config(num_tokens, num_experts,
+                                        shard_intermediate_size, hidden_size,
+                                        topk, dtype_str)
+        else:
+            config = op_config[min(op_config.keys(),
+                                   key=lambda x: abs(x - num_tokens))]
+        kernel_time = benchmark_config(config, num_tokens, num_experts,
+                                       shard_intermediate_size, hidden_size,
+                                       topk, dtype, use_fp8)
+        return config, kernel_time
+
+    def tune(
+        self,
+        num_tokens: int,
+        num_experts: int,
+        shard_intermediate_size: int,
+        hidden_size: int,
+        topk: int,
+        dtype: torch.dtype,
+        use_fp8: bool,
+        search_space: List[Dict[str, int]],
+    ) -> Dict[str, int]:
+        best_config = None
+        best_time = float("inf")
+        for config in tqdm(search_space):
+            try:
+                kernel_time = benchmark_config(config,
+                                               num_tokens,
+                                               num_experts,
+                                               shard_intermediate_size,
+                                               hidden_size,
+                                               topk,
+                                               dtype,
+                                               use_fp8,
+                                               num_iters=10)
+            except triton.runtime.autotuner.OutOfResources:
+                # Some configurations may be invalid and fail to compile.
+                continue
+
+            if kernel_time < best_time:
+                best_time = kernel_time
+                best_config = config
+        now = datetime.now()
+        print(f"{now.ctime()}] Completed tuning for batch_size={num_tokens}")
+        return best_config
+
+
+def sort_config(config: Dict[str, int]) -> Dict[str, int]:
+    return {
+        "BLOCK_SIZE_M": config["BLOCK_SIZE_M"],
+        "BLOCK_SIZE_N": config["BLOCK_SIZE_N"],
+        "BLOCK_SIZE_K": config["BLOCK_SIZE_K"],
+        "GROUP_SIZE_M": config["GROUP_SIZE_M"],
+        "num_warps": config["num_warps"],
+        "num_stages": config["num_stages"],
+    }
+
+
+def save_configs(
+    configs: Dict[int, Dict[str, int]],
+    num_experts: int,
+    shard_intermediate_size: int,
+    hidden_size: int,
+    topk: int,
+    dtype: torch.dtype,
+    use_fp8: bool,
+) -> None:
+    dtype_str = "float8" if use_fp8 else None
+    # NOTE(woosuk): The current naming convention uses w2.shape[2], which
+    # is the intermediate size after silu_and_mul.
+    filename = get_config_file_name(num_experts, shard_intermediate_size // 2,
+                                    dtype_str)
+    print(f"Writing best config to {filename}...")
+    with open(filename, "w") as f:
+        json.dump(configs, f, indent=4)
+        f.write("\n")
+
+
+def main(args: argparse.Namespace):
+    print(args)
+
+    config = AutoConfig.from_pretrained(args.model)
+    if config.architectures[0] == "DbrxForCausalLM":
+        E = config.ffn_config.moe_num_experts
+        topk = config.ffn_config.moe_top_k
+        intermediate_size = config.ffn_config.ffn_hidden_size
+        shard_intermediate_size = 2 * intermediate_size // args.tp_size
+    else:
+        # Default: Mixtral.
+        E = config.num_local_experts
+        topk = config.num_experts_per_tok
+        intermediate_size = config.intermediate_size
+        shard_intermediate_size = 2 * intermediate_size // args.tp_size
+
+    hidden_size = config.hidden_size
+    dtype = config.torch_dtype
+    use_fp8 = args.dtype == "fp8"
+
+    if args.batch_size is None:
+        batch_sizes = [
+            1, 2, 4, 8, 16, 24, 32, 48, 64, 96, 128, 256, 512, 1024, 1536,
+            2048, 3072, 4096
+        ]
+    else:
+        batch_sizes = [args.batch_size]
+
+    ray.init()
+    num_gpus = int(ray.available_resources()["GPU"])
+    workers = [BenchmarkWorker.remote(args.seed) for _ in range(num_gpus)]
+
+    def _distribute(method: str, inputs: List[Any]) -> List[Any]:
+        outputs = []
+        worker_idx = 0
+        for input_args in inputs:
+            worker = workers[worker_idx]
+            worker_method = getattr(worker, method)
+            output = worker_method.remote(*input_args)
+            outputs.append(output)
+            worker_idx = (worker_idx + 1) % num_gpus
+        return ray.get(outputs)
+
+    if args.tune:
+        search_space = get_configs_compute_bound()
+        print(f"Start tuning over {len(search_space)} configurations...")
+
+        start = time.time()
+        configs = _distribute(
+            "tune", [(batch_size, E, shard_intermediate_size, hidden_size,
+                      topk, dtype, use_fp8, search_space)
+                     for batch_size in batch_sizes])
+        best_configs = {
+            M: sort_config(config)
+            for M, config in zip(batch_sizes, configs)
+        }
+        save_configs(best_configs, E, shard_intermediate_size, hidden_size,
+                     topk, dtype, use_fp8)
+        end = time.time()
+        print(f"Tuning took {end - start:.2f} seconds")
+    else:
+        outputs = _distribute("benchmark",
+                              [(batch_size, E, shard_intermediate_size,
+                                hidden_size, topk, dtype, use_fp8)
+                               for batch_size in batch_sizes])
+
+        for batch_size, (config, kernel_time) in zip(batch_sizes, outputs):
+            print(f"Batch size: {batch_size}, config: {config}")
+            print(f"Kernel time: {kernel_time:.2f} us")
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser()
+    parser.add_argument("--model",
+                        type=str,
+                        default="mistralai/Mixtral-8x7B-Instruct-v0.1")
+    parser.add_argument("--tp-size", "-tp", type=int, default=2)
+    parser.add_argument("--dtype",
+                        type=str,
+                        choices=["auto", "fp8"],
+                        default="auto")
+    parser.add_argument("--seed", type=int, default=0)
+    parser.add_argument("--batch-size", type=int, required=False)
+    parser.add_argument("--tune", action="store_true")
+    args = parser.parse_args()
+
+    main(args)

benchmarks/kernels/benchmark_paged_attention.py

@@ -16,7 +16,7 @@ PARTITION_SIZE = 512
 def main(
     version: str,
     num_seqs: int,
-    context_len: int,
+    seq_len: int,
     num_query_heads: int,
     num_kv_heads: int,
     head_size: int,
@@ -48,12 +48,12 @@ def main(
                                    dtype=torch.float,
                                    device=device)
 
-    context_lens = [context_len for _ in range(num_seqs)]
+    seq_lens = [seq_len for _ in range(num_seqs)]
-    max_context_len = max(context_lens)
+    max_seq_len = max(seq_lens)
-    context_lens = torch.tensor(context_lens, dtype=torch.int, device=device)
+    seq_lens = torch.tensor(seq_lens, dtype=torch.int, device=device)
 
     # Create the block tables.
-    max_num_blocks_per_seq = (max_context_len + block_size - 1) // block_size
+    max_num_blocks_per_seq = (max_seq_len + block_size - 1) // block_size
     block_tables = []
     for _ in range(num_seqs):
         block_table = [
@@ -77,8 +77,7 @@ def main(
     # Prepare for the paged attention kernel.
     output = torch.empty_like(query)
     if version == "v2":
-        num_partitions = ((max_context_len + PARTITION_SIZE - 1) //
+        num_partitions = ((max_seq_len + PARTITION_SIZE - 1) // PARTITION_SIZE)
-                          PARTITION_SIZE)
         tmp_output = torch.empty(
             size=(num_seqs, num_query_heads, num_partitions, head_size),
             dtype=output.dtype,
@@ -110,9 +109,9 @@ def main(
                     num_kv_heads,
                     scale,
                     block_tables,
-                    context_lens,
+                    seq_lens,
                     block_size,
-                    max_context_len,
+                    max_seq_len,
                     alibi_slopes,
                     kv_cache_dtype,
                     kv_scale,
@@ -129,9 +128,9 @@ def main(
                     num_kv_heads,
                     scale,
                     block_tables,
-                    context_lens,
+                    seq_lens,
                     block_size,
-                    max_context_len,
+                    max_seq_len,
                     alibi_slopes,
                     kv_cache_dtype,
                     kv_scale,
@@ -166,12 +165,12 @@ if __name__ == '__main__':
                         choices=["v1", "v2"],
                         default="v2")
     parser.add_argument("--batch-size", type=int, default=8)
-    parser.add_argument("--context-len", type=int, default=4096)
+    parser.add_argument("--seq_len", type=int, default=4096)
     parser.add_argument("--num-query-heads", type=int, default=64)
     parser.add_argument("--num-kv-heads", type=int, default=8)
     parser.add_argument("--head-size",
                         type=int,
-                        choices=[64, 80, 96, 112, 128, 256],
+                        choices=[64, 80, 96, 112, 128, 192, 256],
                         default=128)
     parser.add_argument("--block-size", type=int, choices=[16, 32], default=16)
     parser.add_argument("--use-alibi", action="store_true")
@@ -184,13 +183,11 @@ if __name__ == '__main__':
     parser.add_argument(
         "--kv-cache-dtype",
         type=str,
-        choices=["auto", "fp8"],
+        choices=["auto", "fp8", "fp8_e5m2", "fp8_e4m3"],
         default="auto",
-        help=
+        help="Data type for kv cache storage. If 'auto', will use model "
-        'Data type for kv cache storage. If "auto", will use model data type. '
+        "data type. CUDA 11.8+ supports fp8 (=fp8_e4m3) and fp8_e5m2. "
-        'FP8_E5M2 (without scaling) is only supported on cuda version greater '
+        "ROCm (AMD GPU) supports fp8 (=fp8_e4m3)")
-        'than 11.8. On ROCm (AMD GPU), FP8_E4M3 is instead supported for '
-        'common inference criteria.')
     args = parser.parse_args()
     print(args)
 
@@ -199,7 +196,7 @@ if __name__ == '__main__':
     main(
         version=args.version,
         num_seqs=args.batch_size,
-        context_len=args.context_len,
+        seq_len=args.seq_len,
         num_query_heads=args.num_query_heads,
         num_kv_heads=args.num_kv_heads,
         head_size=args.head_size,

benchmarks/kernels/benchmark_rope.py

@@ -93,7 +93,7 @@ if __name__ == '__main__':
     parser.add_argument("--num-heads", type=int, default=8)
     parser.add_argument("--head-size",
                         type=int,
-                        choices=[64, 80, 96, 112, 128, 256],
+                        choices=[64, 80, 96, 112, 128, 192, 256],
                         default=128)
     parser.add_argument("--rotary-dim", type=int, choices=[16, 32], default=32)
     parser.add_argument("--dtype",

benchmarks/kernels/benchmark_shapes.py

@@ -0,0 +1,75 @@
+WEIGHT_SHAPES = {
+    "ideal": [[4 * 256 * 32, 256 * 32]],
+    "mistralai/Mistral-7B-v0.1/TP1": [
+        [4096, 6144],
+        [4096, 4096],
+        [4096, 28672],
+        [14336, 4096],
+    ],
+    "mistralai/Mistral-7B-v0.1/TP2": [
+        [4096, 3072],
+        [2048, 4096],
+        [4096, 14336],
+        [7168, 4096],
+    ],
+    "mistralai/Mistral-7B-v0.1/TP4": [
+        [4096, 1536],
+        [1024, 4096],
+        [4096, 7168],
+        [3584, 4096],
+    ],
+    "meta-llama/Llama-2-7b-hf/TP1": [
+        [4096, 12288],
+        [4096, 4096],
+        [4096, 22016],
+        [11008, 4096],
+    ],
+    "meta-llama/Llama-2-7b-hf/TP2": [
+        [4096, 6144],
+        [2048, 4096],
+        [4096, 11008],
+        [5504, 4096],
+    ],
+    "meta-llama/Llama-2-7b-hf/TP4": [
+        [4096, 3072],
+        [1024, 4096],
+        [4096, 5504],
+        [2752, 4096],
+    ],
+    "meta-llama/Llama-2-13b-hf/TP1": [
+        [5120, 15360],
+        [5120, 5120],
+        [5120, 27648],
+        [13824, 5120],
+    ],
+    "meta-llama/Llama-2-13b-hf/TP2": [
+        [5120, 7680],
+        [2560, 5120],
+        [5120, 13824],
+        [6912, 5120],
+    ],
+    "meta-llama/Llama-2-13b-hf/TP4": [
+        [5120, 3840],
+        [1280, 5120],
+        [5120, 6912],
+        [3456, 5120],
+    ],
+    "meta-llama/Llama-2-70b-hf/TP1": [
+        [8192, 10240],
+        [8192, 8192],
+        [8192, 57344],
+        [28672, 8192],
+    ],
+    "meta-llama/Llama-2-70b-hf/TP2": [
+        [8192, 5120],
+        [4096, 8192],
+        [8192, 28672],
+        [14336, 8192],
+    ],
+    "meta-llama/Llama-2-70b-hf/TP4": [
+        [8192, 2560],
+        [2048, 8192],
+        [8192, 14336],
+        [7168, 8192],
+    ],
+}

benchmarks/launch_tgi_server.sh

@@ -4,7 +4,7 @@ PORT=8000
 MODEL=$1
 TOKENS=$2
 
-docker run --gpus all --shm-size 1g -p $PORT:80 \
+docker run -e HF_TOKEN=$HF_TOKEN --gpus all --shm-size 1g -p $PORT:80 \
            -v $PWD/data:/data \
            ghcr.io/huggingface/text-generation-inference:1.4.0 \
            --model-id $MODEL \

benchmarks/overheads/benchmark_hashing.py

@@ -0,0 +1,63 @@
+import argparse
+import cProfile
+import pstats
+
+from vllm import LLM, SamplingParams
+
+# A very long prompt, total number of tokens is about 15k.
+LONG_PROMPT = ["You are an expert in large language models, aren't you?"
+               ] * 1000
+LONG_PROMPT = ' '.join(LONG_PROMPT)
+
+
+def main(args):
+    llm = LLM(
+        model=args.model,
+        enforce_eager=True,
+        enable_prefix_caching=True,
+        tensor_parallel_size=args.tensor_parallel_size,
+        use_v2_block_manager=args.use_v2_block_manager,
+    )
+
+    sampling_params = SamplingParams(temperature=0, max_tokens=args.output_len)
+    profiler = cProfile.Profile()
+
+    print("------warm up------")
+    for i in range(3):
+        output = llm.generate(LONG_PROMPT, sampling_params)
+        print(output[0].outputs[0].text)
+
+    print("------start generating------")
+    for i in range(3):
+        profiler.runctx('llm.generate(LONG_PROMPT, sampling_params)',
+                        globals(), locals())
+
+    # analyze the runtime of hashing function
+    stats = pstats.Stats(profiler)
+    stats.sort_stats('cumulative')
+    total_time = 0
+    total_calls = 0
+    for func in stats.stats:
+        if 'hash_of_block' in func[2]:
+            total_time = stats.stats[func][3]
+            total_calls = stats.stats[func][0]
+    percentage = (total_time / stats.total_tt) * 100
+    print(f"Hashing took {total_time:.2f} seconds,"
+          f"{percentage:.2f}% of the total runtime.")
+
+
+if __name__ == "__main__":
+    parser = argparse.ArgumentParser(
+        description='Benchmark the performance of hashing function in'
+        'automatic prefix caching.')
+    parser.add_argument('--model', type=str, default='lmsys/longchat-7b-16k')
+    parser.add_argument('--tensor-parallel-size', '-tp', type=int, default=1)
+    parser.add_argument('--output-len', type=int, default=10)
+    parser.add_argument('--enable-prefix-caching',
+                        action='store_true',
+                        help='enable prefix caching')
+    parser.add_argument('--use-v2-block-manager',
+                        action='store_true',
+                        help='Use BlockSpaceMangerV2')
+    args = parser.parse_args()
+    main(args)

cmake/cpu_extension.cmake

@@ -73,7 +73,7 @@ set(VLLM_EXT_SRC
     "csrc/cpu/cache.cpp"
     "csrc/cpu/layernorm.cpp"
     "csrc/cpu/pos_encoding.cpp"
-    "csrc/cpu/pybind.cpp")
+    "csrc/cpu/torch_bindings.cpp")
 
 define_gpu_extension_target(
     _C
@@ -81,10 +81,10 @@ define_gpu_extension_target(
     LANGUAGE CXX
     SOURCES ${VLLM_EXT_SRC}
     COMPILE_FLAGS ${CXX_COMPILE_FLAGS}
+    USE_SABI 3
     WITH_SOABI
 )
 
 add_custom_target(default)
 message(STATUS "Enabling C extension.")
 add_dependencies(default _C)
-

cmake/utils.cmake

@@ -5,7 +5,7 @@
 macro (find_python_from_executable EXECUTABLE SUPPORTED_VERSIONS)
   file(REAL_PATH ${EXECUTABLE} EXECUTABLE)
   set(Python_EXECUTABLE ${EXECUTABLE})
-  find_package(Python COMPONENTS Interpreter Development.Module)
+  find_package(Python COMPONENTS Interpreter Development.Module Development.SABIModule)
   if (NOT Python_FOUND)
     message(FATAL_ERROR "Unable to find python matching: ${EXECUTABLE}.")
   endif()
@@ -99,7 +99,7 @@ function (get_torch_gpu_compiler_flags OUT_GPU_FLAGS GPU_LANG)
       "Failed to determine torch nvcc compiler flags")
 
     if (CUDA_VERSION VERSION_GREATER_EQUAL 11.8)
-      list(APPEND GPU_FLAGS "-DENABLE_FP8_E5M2")
+      list(APPEND GPU_FLAGS "-DENABLE_FP8")
     endif()
     if (CUDA_VERSION VERSION_GREATER_EQUAL 12.0)
       list(REMOVE_ITEM GPU_FLAGS
@@ -119,7 +119,7 @@ function (get_torch_gpu_compiler_flags OUT_GPU_FLAGS GPU_LANG)
 
     list(APPEND GPU_FLAGS
       "-DUSE_ROCM"
-      "-DENABLE_FP8_E4M3"
+      "-DENABLE_FP8"
       "-U__HIP_NO_HALF_CONVERSIONS__"
       "-U__HIP_NO_HALF_OPERATORS__"
       "-fno-gpu-rdc")
@@ -294,6 +294,7 @@ endmacro()
 # INCLUDE_DIRECTORIES <dirs> - Extra include directories.
 # LIBRARIES <libraries>      - Extra link libraries.
 # WITH_SOABI                 - Generate library with python SOABI suffix name.
+# USE_SABI <version>         - Use python stable api <version>
 #
 # Note: optimization level/debug info is set via cmake build type.
 #
@@ -301,7 +302,7 @@ function (define_gpu_extension_target GPU_MOD_NAME)
   cmake_parse_arguments(PARSE_ARGV 1
     GPU
     "WITH_SOABI"
-    "DESTINATION;LANGUAGE"
+    "DESTINATION;LANGUAGE;USE_SABI"
     "SOURCES;ARCHITECTURES;COMPILE_FLAGS;INCLUDE_DIRECTORIES;LIBRARIES")
 
   # Add hipify preprocessing step when building with HIP/ROCm.
@@ -315,7 +316,11 @@ function (define_gpu_extension_target GPU_MOD_NAME)
     set(GPU_WITH_SOABI)
   endif()
 
-  Python_add_library(${GPU_MOD_NAME} MODULE "${GPU_SOURCES}" ${GPU_WITH_SOABI})
+  if (GPU_USE_SABI)
+    Python_add_library(${GPU_MOD_NAME} MODULE USE_SABI ${GPU_USE_SABI} ${GPU_WITH_SOABI} "${GPU_SOURCES}")
+  else()
+    Python_add_library(${GPU_MOD_NAME} MODULE ${GPU_WITH_SOABI} "${GPU_SOURCES}")
+  endif()
 
   if (GPU_LANGUAGE STREQUAL "HIP")
     # Make this target dependent on the hipify preprocessor step.

collect_env.py

@@ -64,6 +64,7 @@ DEFAULT_CONDA_PATTERNS = {
     "triton",
     "optree",
     "nccl",
+    "transformers",
 }
 
 DEFAULT_PIP_PATTERNS = {
@@ -75,6 +76,7 @@ DEFAULT_PIP_PATTERNS = {
     "optree",
     "onnx",
     "nccl",
+    "transformers",
 }
 
 
@@ -601,6 +603,11 @@ Versions of relevant libraries:
 {conda_packages}
 """.strip()
 
+# both the above code and the following code use `strip()` to
+# remove leading/trailing whitespaces, so we need to add a newline
+# in between to separate the two sections
+env_info_fmt += "\n"
+
 env_info_fmt += """
 ROCM Version: {rocm_version}
 Neuron SDK Version: {neuron_sdk_version}

csrc/activation_kernels.cu

@@ -1,5 +1,5 @@
 #include <ATen/cuda/CUDAContext.h>
-#include <torch/extension.h>
+#include <torch/all.h>
 #include <c10/cuda/CUDAGuard.h>
 
 #include <cmath>
@@ -63,31 +63,25 @@ __device__ __forceinline__ T gelu_tanh_kernel(const T& x) {
   const at::cuda::OptionalCUDAGuard device_guard(device_of(input));      \
   const cudaStream_t stream = at::cuda::getCurrentCUDAStream();          \
   VLLM_DISPATCH_FLOATING_TYPES(                                          \
-    input.scalar_type(),                                                                  \
+      input.scalar_type(), "act_and_mul_kernel", [&] {                   \
-    "act_and_mul_kernel",                                                                 \
+        vllm::act_and_mul_kernel<scalar_t, KERNEL<scalar_t>>             \
-    [&] {                                                                                 \
+            <<<grid, block, 0, stream>>>(out.data_ptr<scalar_t>(),       \
-      vllm::act_and_mul_kernel<scalar_t, KERNEL<scalar_t>><<<grid, block, 0, stream>>>(   \
+                                         input.data_ptr<scalar_t>(), d); \
-        out.data_ptr<scalar_t>(),                                                         \
-        input.data_ptr<scalar_t>(),                                                       \
-        d);                                                                               \
       });
 
-void silu_and_mul(
+void silu_and_mul(torch::Tensor& out,    // [..., d]
-  torch::Tensor& out,      // [..., d]
                   torch::Tensor& input)  // [..., 2 * d]
 {
   LAUNCH_ACTIVATION_GATE_KERNEL(vllm::silu_kernel);
 }
 
-void gelu_and_mul(
+void gelu_and_mul(torch::Tensor& out,    // [..., d]
-  torch::Tensor& out,      // [..., d]
                   torch::Tensor& input)  // [..., 2 * d]
 {
   LAUNCH_ACTIVATION_GATE_KERNEL(vllm::gelu_kernel);
 }
 
-void gelu_tanh_and_mul(
+void gelu_tanh_and_mul(torch::Tensor& out,    // [..., d]
-  torch::Tensor& out,      // [..., d]
                        torch::Tensor& input)  // [..., 2 * d]
 {
   LAUNCH_ACTIVATION_GATE_KERNEL(vllm::gelu_tanh_kernel);
@@ -118,14 +112,10 @@ __global__ void activation_kernel(
   dim3 block(std::min(d, 1024));                                               \
   const at::cuda::OptionalCUDAGuard device_guard(device_of(input));            \
   const cudaStream_t stream = at::cuda::getCurrentCUDAStream();                \
-  VLLM_DISPATCH_FLOATING_TYPES(                                                           \
+  VLLM_DISPATCH_FLOATING_TYPES(input.scalar_type(), "activation_kernel", [&] { \
-    input.scalar_type(),                                                                  \
+    vllm::activation_kernel<scalar_t, KERNEL<scalar_t>>                        \
-    "activation_kernel",                                                                  \
+        <<<grid, block, 0, stream>>>(out.data_ptr<scalar_t>(),                 \
-    [&] {                                                                                 \
+                                     input.data_ptr<scalar_t>(), d);           \
-      vllm::activation_kernel<scalar_t, KERNEL<scalar_t>><<<grid, block, 0, stream>>>(    \
-        out.data_ptr<scalar_t>(),                                                         \
-        input.data_ptr<scalar_t>(),                                                       \
-        d);                                                                               \
   });
 
 namespace vllm {
@@ -140,21 +130,20 @@ __device__ __forceinline__ T gelu_new_kernel(const T& x) {
 template <typename T>
 __device__ __forceinline__ T gelu_fast_kernel(const T& x) {
   const float f = (float)x;
-  const T t = (T) tanhf(((T) (f * 0.79788456f)) * (((T) 1.0) + (T) (0.044715f * f) * x));
+  const T t =
+      (T)tanhf(((T)(f * 0.79788456f)) * (((T)1.0) + (T)(0.044715f * f) * x));
   return ((T)0.5) * x * (((T)1.0) + t);
 }
 
 }  // namespace vllm
 
-void gelu_new(
+void gelu_new(torch::Tensor& out,    // [..., d]
-  torch::Tensor& out,     // [..., d]
               torch::Tensor& input)  // [..., d]
 {
   LAUNCH_ACTIVATION_KERNEL(vllm::gelu_new_kernel);
 }
 
-void gelu_fast(
+void gelu_fast(torch::Tensor& out,    // [..., d]
-  torch::Tensor& out,     // [..., d]
                torch::Tensor& input)  // [..., d]
 {
   LAUNCH_ACTIVATION_KERNEL(vllm::gelu_fast_kernel);

csrc/attention/attention_generic.cuh

@@ -1,5 +1,6 @@
 /*
- * Adapted from https://github.com/NVIDIA/FasterTransformer/blob/release/v5.3_tag/src/fastertransformer/kernels/decoder_masked_multihead_attention_utils.h
+ * Adapted from
+ * https://github.com/NVIDIA/FasterTransformer/blob/release/v5.3_tag/src/fastertransformer/kernels/decoder_masked_multihead_attention_utils.h
  * Copyright (c) 2023, The vLLM team.
  * Copyright (c) 2020-2023, NVIDIA CORPORATION.  All rights reserved.
  *

csrc/attention/attention_utils.cuh

@@ -1,5 +1,6 @@
 /*
- * Adapted from https://github.com/NVIDIA/FasterTransformer/blob/release/v5.3_tag/src/fastertransformer/kernels/decoder_masked_multihead_attention/decoder_masked_multihead_attention_template.hpp
+ * Adapted from
+ * https://github.com/NVIDIA/FasterTransformer/blob/release/v5.3_tag/src/fastertransformer/kernels/decoder_masked_multihead_attention/decoder_masked_multihead_attention_template.hpp
  * Copyright (c) 2023, The vLLM team.
  * Copyright (c) 2020-2023, NVIDIA CORPORATION.  All rights reserved.
  *

csrc/attention/dtype_bfloat16.cuh

@@ -1,6 +1,8 @@
 /*
- * Adapted from https://github.com/NVIDIA/FasterTransformer/blob/release/v5.3_tag/src/fastertransformer/kernels/decoder_masked_multihead_attention/decoder_masked_multihead_attention_template.hpp
+ * Adapted from
- * and https://github.com/NVIDIA/FasterTransformer/blob/release/v5.3_tag/src/fastertransformer/kernels/decoder_masked_multihead_attention_utils.h
+ * https://github.com/NVIDIA/FasterTransformer/blob/release/v5.3_tag/src/fastertransformer/kernels/decoder_masked_multihead_attention/decoder_masked_multihead_attention_template.hpp
+ * and
+ * https://github.com/NVIDIA/FasterTransformer/blob/release/v5.3_tag/src/fastertransformer/kernels/decoder_masked_multihead_attention_utils.h
  * Copyright (c) 2023, The vLLM team.
  * Copyright (c) 2020-2023, NVIDIA CORPORATION.  All rights reserved.
  *
@@ -280,7 +282,8 @@ inline __device__ Float8_ mul(__nv_bfloat16 a, bf16_8_t b) {
 }
 
 // Vector fused multiply-add.
-inline __device__ __nv_bfloat162 fma(__nv_bfloat162 a, __nv_bfloat162 b, __nv_bfloat162 c) {
+inline __device__ __nv_bfloat162 fma(__nv_bfloat162 a, __nv_bfloat162 b,
+                                     __nv_bfloat162 c) {
 #if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 800
   assert(false);
 #else
@@ -288,7 +291,8 @@ inline __device__ __nv_bfloat162 fma(__nv_bfloat162 a, __nv_bfloat162 b, __nv_bf
 #endif
 }
 
-inline __device__ __nv_bfloat162 fma(__nv_bfloat16 a, __nv_bfloat162 b, __nv_bfloat162 c) {
+inline __device__ __nv_bfloat162 fma(__nv_bfloat16 a, __nv_bfloat162 b,
+                                     __nv_bfloat162 c) {
 #if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 800
   assert(false);
 #else

csrc/attention/dtype_float16.cuh

@@ -1,6 +1,8 @@
 /*
- * Adapted from https://github.com/NVIDIA/FasterTransformer/blob/release/v5.3_tag/src/fastertransformer/kernels/decoder_masked_multihead_attention/decoder_masked_multihead_attention_template.hpp
+ * Adapted from
- * and https://github.com/NVIDIA/FasterTransformer/blob/release/v5.3_tag/src/fastertransformer/kernels/decoder_masked_multihead_attention_utils.h
+ * https://github.com/NVIDIA/FasterTransformer/blob/release/v5.3_tag/src/fastertransformer/kernels/decoder_masked_multihead_attention/decoder_masked_multihead_attention_template.hpp
+ * and
+ * https://github.com/NVIDIA/FasterTransformer/blob/release/v5.3_tag/src/fastertransformer/kernels/decoder_masked_multihead_attention_utils.h
  * Copyright (c) 2023, The vLLM team.
  * Copyright (c) 2020-2023, NVIDIA CORPORATION.  All rights reserved.
  *
@@ -130,7 +132,9 @@ inline __device__ uint32_t float2_to_half2(float2 f) {
   } tmp;
 #ifndef USE_ROCM
   #if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 800
-    asm volatile("cvt.rn.f16x2.f32 %0, %1, %2;\n" : "=r"(tmp.u32) : "f"(f.y), "f"(f.x));
+  asm volatile("cvt.rn.f16x2.f32 %0, %1, %2;\n"
+               : "=r"(tmp.u32)
+               : "f"(f.y), "f"(f.x));
   #else
   asm volatile("cvt.rn.f16.f32 %0, %1;\n" : "=h"(tmp.u16[0]) : "f"(f.x));
   asm volatile("cvt.rn.f16.f32 %0, %1;\n" : "=h"(tmp.u16[1]) : "f"(f.y));
@@ -327,9 +331,13 @@ inline __device__ Float8_ mul(uint16_t a, uint4 b) {
 inline __device__ uint32_t fma(uint32_t a, uint32_t b, uint32_t c) {
   uint32_t d;
 #ifndef USE_ROCM
-  asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n" : "=r"(d) : "r"(a), "r"(b), "r"(c));
+  asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n"
+               : "=r"(d)
+               : "r"(a), "r"(b), "r"(c));
 #else
-  asm volatile("v_pk_fma_f16 %0, %1, %2, %3;\n" : "=v"(d) : "v"(a), "v"(b), "v"(c));
+  asm volatile("v_pk_fma_f16 %0, %1, %2, %3;\n"
+               : "=v"(d)
+               : "v"(a), "v"(b), "v"(c));
 #endif
   return d;
 }
@@ -470,13 +478,9 @@ inline __device__ void from_float(uint4& dst, Float8_ src) {
 }
 
 // From float16 to float32.
-inline __device__ float to_float(uint16_t u) {
+inline __device__ float to_float(uint16_t u) { return half_to_float(u); }
-  return half_to_float(u);
-}
 
-inline __device__ float2 to_float(uint32_t u) {
+inline __device__ float2 to_float(uint32_t u) { return half2_to_float2(u); }
-  return half2_to_float2(u);
-}
 
 inline __device__ Float4_ to_float(uint2 u) {
   Float4_ tmp;
@@ -495,8 +499,6 @@ inline __device__ Float8_ to_float(uint4 u) {
 }
 
 // Zero-out a variable.
-inline __device__ void zero(uint16_t& dst) {
+inline __device__ void zero(uint16_t& dst) { dst = uint16_t(0); }
-  dst = uint16_t(0);
-}
 
 }  // namespace vllm

csrc/attention/dtype_float32.cuh

@@ -1,6 +1,8 @@
 /*
- * Adapted from https://github.com/NVIDIA/FasterTransformer/blob/release/v5.3_tag/src/fastertransformer/kernels/decoder_masked_multihead_attention/decoder_masked_multihead_attention_template.hpp
+ * Adapted from
- * and https://github.com/NVIDIA/FasterTransformer/blob/release/v5.3_tag/src/fastertransformer/kernels/decoder_masked_multihead_attention_utils.h
+ * https://github.com/NVIDIA/FasterTransformer/blob/release/v5.3_tag/src/fastertransformer/kernels/decoder_masked_multihead_attention/decoder_masked_multihead_attention_template.hpp
+ * and
+ * https://github.com/NVIDIA/FasterTransformer/blob/release/v5.3_tag/src/fastertransformer/kernels/decoder_masked_multihead_attention_utils.h
  * Copyright (c) 2023, The vLLM team.
  * Copyright (c) 2020-2023, NVIDIA CORPORATION.  All rights reserved.
  *
@@ -66,9 +68,7 @@ struct FloatVec<float4> {
 };
 
 // Vector addition.
-inline __device__ float add(float a, float b) {
+inline __device__ float add(float a, float b) { return a + b; }
-  return a + b;
-}
 
 inline __device__ float2 add(float2 a, float2 b) {
   float2 c;
@@ -129,9 +129,7 @@ inline __device__ float4 mul(float a, float4 b) {
 }
 
 // Vector fused multiply-add.
-inline __device__ float fma(float a, float b, float c) {
+inline __device__ float fma(float a, float b, float c) { return a * b + c; }
-  return a * b + c;
-}
 
 inline __device__ float2 fma(float2 a, float2 b, float2 c) {
   float2 d;
@@ -208,9 +206,7 @@ inline __device__ float sum(Float8_ v) {
 }
 
 // Vector dot product.
-inline __device__ float dot(float a, float b) {
+inline __device__ float dot(float a, float b) { return a * b; }
-  return a * b;
-}
 
 inline __device__ float dot(float2 a, float2 b) {
   float2 c = mul<float2, float2, float2>(a, b);
@@ -232,42 +228,24 @@ inline __device__ float dot(Float8_ a, Float8_ b) {
 }
 
 // From float to float.
-inline __device__ void from_float(float& dst, float src) {
+inline __device__ void from_float(float& dst, float src) { dst = src; }
-  dst = src;
-}
 
-inline __device__ void from_float(float2& dst, float2 src) {
+inline __device__ void from_float(float2& dst, float2 src) { dst = src; }
-  dst = src;
-}
 
-inline __device__ void from_float(float4& dst, float4 src) {
+inline __device__ void from_float(float4& dst, float4 src) { dst = src; }
-  dst = src;
-}
 
 // From float to float.
-inline __device__ float to_float(float u) {
+inline __device__ float to_float(float u) { return u; }
-  return u;
-}
 
-inline __device__ float2 to_float(float2 u) {
+inline __device__ float2 to_float(float2 u) { return u; }
-  return u;
-}
 
-inline __device__ float4 to_float(float4 u) {
+inline __device__ float4 to_float(float4 u) { return u; }
-  return u;
-}
 
-inline __device__ Float4_ to_float(Float4_ u) {
+inline __device__ Float4_ to_float(Float4_ u) { return u; }
-  return u;
-}
 
-inline __device__ Float8_ to_float(Float8_ u) {
+inline __device__ Float8_ to_float(Float8_ u) { return u; }
-  return u;
-}
 
 // Zero-out a variable.
-inline __device__ void zero(float& dst) {
+inline __device__ void zero(float& dst) { dst = 0.f; }
-  dst = 0.f;
-}
 
 }  // namespace vllm

csrc/attention/dtype_fp8.cuh

@@ -3,14 +3,21 @@
 #include "attention_generic.cuh"
 
 #include <stdint.h>
-#ifdef ENABLE_FP8_E5M2
+#ifdef ENABLE_FP8
+  #ifndef USE_ROCM
     #include <cuda_fp8.h>
-#endif
+  #endif  // USE_ROCM
+#endif    // ENABLE_FP8
 
 namespace vllm {
-#if defined(ENABLE_FP8_E5M2) || defined(ENABLE_FP8_E4M3)
-// fp8 vector types for quantization of kv cache
 
+enum class Fp8KVCacheDataType {
+  kAuto = 0,
+  kFp8E4M3 = 1,
+  kFp8E5M2 = 2,
+};
+
+// fp8 vector types for quantization of kv cache
 template <>
 struct Vec<uint8_t, 1> {
   using Type = uint8_t;
@@ -30,6 +37,5 @@ template<>
 struct Vec<uint8_t, 8> {
   using Type = uint2;
 };
-#endif // ENABLE_FP8_E5M2
 
 }  // namespace vllm

csrc/cache.h

@@ -1,30 +1,32 @@
 #pragma once
 
-#include <torch/extension.h>
+#include <torch/all.h>
 
 #include <map>
 #include <vector>
 
-void swap_blocks(
+void swap_blocks(torch::Tensor& src, torch::Tensor& dst,
-  torch::Tensor& src,
+                 const torch::Tensor& block_mapping);
-  torch::Tensor& dst,
-  const std::map<int64_t, int64_t>& block_mapping);
 
-void copy_blocks(
+// Note: the key_caches and value_caches vectors are constant but
-  std::vector<torch::Tensor>& key_caches,
+// not the Tensors they contain. The vectors need to be const refs
-  std::vector<torch::Tensor>& value_caches,
+// in order to satisfy pytorch's C++ operator registration code.
-  const std::map<int64_t, std::vector<int64_t>>& block_mapping);
+void copy_blocks(std::vector<torch::Tensor> const& key_caches,
+                 std::vector<torch::Tensor> const& value_caches,
+                 const torch::Tensor& block_mapping);
 
-void reshape_and_cache(
+void reshape_and_cache(torch::Tensor& key, torch::Tensor& value,
-  torch::Tensor& key,
+                       torch::Tensor& key_cache, torch::Tensor& value_cache,
-  torch::Tensor& value,
+                       torch::Tensor& slot_mapping,
+                       const std::string& kv_cache_dtype,
+                       const double kv_scale);
+
+void reshape_and_cache_flash(torch::Tensor& key, torch::Tensor& value,
                              torch::Tensor& key_cache,
                              torch::Tensor& value_cache,
                              torch::Tensor& slot_mapping,
-  const std::string& kv_cache_dtype,
+                             const std::string& kv_cache_dtype);
-  const float kv_scale);
 
 // Just for unittest
-void convert_fp8(
+void convert_fp8(torch::Tensor& dst_cache, torch::Tensor& src_cache,
-  torch::Tensor& src_cache,
+                 const double scale, const std::string& kv_cache_dtype);
-  torch::Tensor& dst_cache);

csrc/cache_kernels.cu

@@ -1,13 +1,14 @@
-#include <torch/extension.h>
+#include <torch/all.h>
 #include <ATen/cuda/CUDAContext.h>
 #include <c10/cuda/CUDAGuard.h>
 
 #include "cuda_compat.h"
 #include "dispatch_utils.h"
-#if defined(ENABLE_FP8_E5M2)
+
-#include "quantization/fp8_e5m2_kvcache/quant_utils.cuh"
+#ifdef USE_ROCM
-#elif defined(ENABLE_FP8_E4M3)
+  #include "quantization/fp8/amd/quant_utils.cuh"
-#include "quantization/fp8/amd_detail/quant_utils.cuh"
+#else
+  #include "quantization/fp8/nvidia/quant_utils.cuh"
 #endif
 
 #include <algorithm>
@@ -20,16 +21,13 @@
 typedef __hip_bfloat16 __nv_bfloat16;
 #endif
 
-void swap_blocks(
+void swap_blocks(torch::Tensor& src, torch::Tensor& dst,
-  torch::Tensor& src,
+                 const torch::Tensor& block_mapping) {
-  torch::Tensor& dst,
-  const std::map<int64_t, int64_t>& block_mapping) {
   torch::Device src_device = src.device();
   torch::Device dst_device = dst.device();
   cudaMemcpyKind memcpy_type;
   if (src_device.is_cuda() && dst_device.is_cuda()) {
-    TORCH_CHECK(
+    TORCH_CHECK(src_device.index() == dst_device.index(),
-      src_device.index() == dst_device.index(),
                 "src and dst must be on the same GPU");
     memcpy_type = cudaMemcpyDeviceToDevice;
   } else if (src_device.is_cuda() && dst_device.is_cpu()) {
@@ -40,24 +38,27 @@ void swap_blocks(
     TORCH_CHECK(false, "Invalid device combination");
   }
 
+  // NOTE(youkaichao): keep in mind that `block_mapping` should be
+  // a cpu tensor, otherwise every `item` call will require a gpu-cpu
+  // synchronization.
+  TORCH_CHECK(block_mapping.device().is_cpu(), "block_mapping must be on CPU");
+
   char* src_ptr = static_cast<char*>(src.data_ptr());
   char* dst_ptr = static_cast<char*>(dst.data_ptr());
 
   const int64_t block_size_in_bytes = src.element_size() * src[0].numel();
-  const at::cuda::OptionalCUDAGuard device_guard(src_device.is_cuda() ? src_device : dst_device);
+  const at::cuda::OptionalCUDAGuard device_guard(
+      src_device.is_cuda() ? src_device : dst_device);
   const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
   // NOTE(woosuk): This can be slow if the number of blocks is large.
-  for (const auto& pair : block_mapping) {
+  const int64_t num_blocks = block_mapping.size(0);
-    int64_t src_block_number = pair.first;
+  for (size_t i = 0; i < num_blocks; i++) {
-    int64_t dst_block_number = pair.second;
+    int64_t src_block_number = block_mapping[i][0].item<int64_t>();
+    int64_t dst_block_number = block_mapping[i][1].item<int64_t>();
     int64_t src_offset = src_block_number * block_size_in_bytes;
     int64_t dst_offset = dst_block_number * block_size_in_bytes;
-    cudaMemcpyAsync(
+    cudaMemcpyAsync(dst_ptr + dst_offset, src_ptr + src_offset,
-      dst_ptr + dst_offset,
+                    block_size_in_bytes, memcpy_type, stream);
-      src_ptr + src_offset,
-      block_size_in_bytes,
-      memcpy_type,
-      stream);
   }
 }
 
@@ -65,8 +66,7 @@ namespace vllm {
 
 // Grid: (num_layers, num_pairs)
 template <typename scalar_t>
-__global__ void copy_blocks_kernel(
+__global__ void copy_blocks_kernel(int64_t* key_cache_ptrs,
-  int64_t* key_cache_ptrs,
                                    int64_t* value_cache_ptrs,
                                    const int64_t* __restrict__ block_mapping,
                                    const int numel_per_block) {
@@ -74,7 +74,8 @@ __global__ void copy_blocks_kernel(
   const int pair_idx = blockIdx.y;
 
   scalar_t* key_cache = reinterpret_cast<scalar_t*>(key_cache_ptrs[layer_idx]);
-  scalar_t* value_cache = reinterpret_cast<scalar_t*>(value_cache_ptrs[layer_idx]);
+  scalar_t* value_cache =
+      reinterpret_cast<scalar_t*>(value_cache_ptrs[layer_idx]);
   int64_t src_block_number = block_mapping[2 * pair_idx];
   int64_t dst_block_number = block_mapping[2 * pair_idx + 1];
 
@@ -94,10 +95,12 @@ __global__ void copy_blocks_kernel(
 
 }  // namespace vllm
 
-void copy_blocks(
+// Note: the key_caches and value_caches vectors are constant but
-  std::vector<torch::Tensor>& key_caches,
+// not the Tensors they contain. The vectors need to be const refs
-  std::vector<torch::Tensor>& value_caches,
+// in order to satisfy pytorch's C++ operator registration code.
-  const std::map<int64_t, std::vector<int64_t>>& block_mapping) {
+void copy_blocks(std::vector<torch::Tensor> const& key_caches,
+                 std::vector<torch::Tensor> const& value_caches,
+                 const torch::Tensor& block_mapping) {
   int num_layers = key_caches.size();
   TORCH_CHECK(num_layers == value_caches.size());
   if (num_layers == 0) {
@@ -111,29 +114,23 @@ void copy_blocks(
   int64_t key_cache_ptrs[num_layers];
   int64_t value_cache_ptrs[num_layers];
   for (int layer_idx = 0; layer_idx < num_layers; ++layer_idx) {
-    key_cache_ptrs[layer_idx] = reinterpret_cast<int64_t>(key_caches[layer_idx].data_ptr());
+    key_cache_ptrs[layer_idx] =
-    value_cache_ptrs[layer_idx] = reinterpret_cast<int64_t>(value_caches[layer_idx].data_ptr());
+        reinterpret_cast<int64_t>(key_caches[layer_idx].data_ptr());
+    value_cache_ptrs[layer_idx] =
+        reinterpret_cast<int64_t>(value_caches[layer_idx].data_ptr());
   }
-  // Create block mapping array.
+
-  std::vector<int64_t> block_mapping_vec;
+  // block_mapping is a 2D tensor with shape (num_pairs, 2).
-  for (const auto& pair : block_mapping) {
+  int num_pairs = block_mapping.size(0);
-    int64_t src_block_number = pair.first;
-    for (int64_t dst_block_number : pair.second) {
-      block_mapping_vec.push_back(src_block_number);
-      block_mapping_vec.push_back(dst_block_number);
-    }
-  }
-  int64_t* block_mapping_array = block_mapping_vec.data();
-  int num_pairs = block_mapping_vec.size() / 2;
 
   // Move the data structures to the GPU.
   // NOTE: This synchronizes the CPU and GPU.
-  torch::Tensor key_cache_ptrs_tensor = torch::from_blob(
+  torch::Tensor key_cache_ptrs_tensor =
-    key_cache_ptrs, {num_layers}, torch::kInt64).to(cache_device);
+      torch::from_blob(key_cache_ptrs, {num_layers}, torch::kInt64)
-  torch::Tensor value_cache_ptrs_tensor = torch::from_blob(
+          .to(cache_device);
-    value_cache_ptrs, {num_layers}, torch::kInt64).to(cache_device);
+  torch::Tensor value_cache_ptrs_tensor =
-  torch::Tensor block_mapping_tensor = torch::from_blob(
+      torch::from_blob(value_cache_ptrs, {num_layers}, torch::kInt64)
-    block_mapping_array, {2 * num_pairs}, torch::kInt64).to(cache_device);
+          .to(cache_device);
 
   // Launch the kernel.
   const int numel_per_block = key_caches[0][0].numel();
@@ -146,26 +143,23 @@ void copy_blocks(
         vllm::copy_blocks_kernel<scalar_t><<<grid, block, 0, stream>>>(
             key_cache_ptrs_tensor.data_ptr<int64_t>(),
             value_cache_ptrs_tensor.data_ptr<int64_t>(),
-        block_mapping_tensor.data_ptr<int64_t>(),
+            block_mapping.data_ptr<int64_t>(), numel_per_block);
-        numel_per_block);
       }));
 }
 
 namespace vllm {
 
-template<typename scalar_t, typename cache_t, bool is_fp8_kv_cache>
+template <typename scalar_t, typename cache_t, Fp8KVCacheDataType kv_dt>
 __global__ void reshape_and_cache_kernel(
     const scalar_t* __restrict__ key,    // [num_tokens, num_heads, head_size]
     const scalar_t* __restrict__ value,  // [num_tokens, num_heads, head_size]
-  cache_t* __restrict__ key_cache,            // [num_blocks, num_heads, head_size/x, block_size, x]
+    cache_t* __restrict__ key_cache,     // [num_blocks, num_heads, head_size/x,
-  cache_t* __restrict__ value_cache,          // [num_blocks, num_heads, head_size, block_size]
+                                         // block_size, x]
+    cache_t* __restrict__ value_cache,   // [num_blocks, num_heads, head_size,
+                                         // block_size]
     const int64_t* __restrict__ slot_mapping,  // [num_tokens]
-  const int key_stride,
+    const int key_stride, const int value_stride, const int num_heads,
-  const int value_stride,
+    const int head_size, const int block_size, const int x,
-  const int num_heads,
-  const int head_size,
-  const int block_size,
-  const int x,
     const float kv_scale) {
   const int64_t token_idx = blockIdx.x;
   const int64_t slot_idx = slot_mapping[token_idx];
@@ -187,60 +181,84 @@ __global__ void reshape_and_cache_kernel(
     const int x_idx = head_offset / x;
     const int x_offset = head_offset % x;
 
-    const int64_t tgt_key_idx = block_idx * num_heads * (head_size / x) * block_size * x
+    const int64_t tgt_key_idx =
-                                + head_idx * (head_size / x) * block_size * x
+        block_idx * num_heads * (head_size / x) * block_size * x +
-                                + x_idx * block_size * x
+        head_idx * (head_size / x) * block_size * x + x_idx * block_size * x +
-                                + block_offset * x
+        block_offset * x + x_offset;
-                                + x_offset;
+    const int64_t tgt_value_idx =
-    const int64_t tgt_value_idx = block_idx * num_heads * head_size * block_size
+        block_idx * num_heads * head_size * block_size +
-                                  + head_idx * head_size * block_size
+        head_idx * head_size * block_size + head_offset * block_size +
-                                  + head_offset * block_size
+        block_offset;
-                                  + block_offset;
     scalar_t tgt_key = key[src_key_idx];
     scalar_t tgt_value = value[src_value_idx];
-    if constexpr (is_fp8_kv_cache) {
+    if constexpr (kv_dt == Fp8KVCacheDataType::kAuto) {
-#if defined(ENABLE_FP8_E5M2)
-      key_cache[tgt_key_idx] = fp8_e5m2_unscaled::vec_conversion<uint8_t, scalar_t>(tgt_key);
-      value_cache[tgt_value_idx] = fp8_e5m2_unscaled::vec_conversion<uint8_t, scalar_t>(tgt_value);
-#elif defined(ENABLE_FP8_E4M3)
-      key_cache[tgt_key_idx] = fp8_e4m3::scaled_vec_conversion<uint8_t, scalar_t>(tgt_key, kv_scale);
-      value_cache[tgt_value_idx] = fp8_e4m3::scaled_vec_conversion<uint8_t, scalar_t>(tgt_value, kv_scale);
-#else
-      assert(false);
-#endif
-    } else {
       key_cache[tgt_key_idx] = tgt_key;
       value_cache[tgt_value_idx] = tgt_value;
+    } else {
+      key_cache[tgt_key_idx] =
+          fp8::scaled_convert<cache_t, scalar_t, kv_dt>(tgt_key, kv_scale);
+      value_cache[tgt_value_idx] =
+          fp8::scaled_convert<cache_t, scalar_t, kv_dt>(tgt_value, kv_scale);
     }
   }
 }
 
+template <typename scalar_t>
+__global__ void reshape_and_cache_flash_kernel(
+    const scalar_t* __restrict__ key,    // [num_tokens, num_heads, head_size]
+    const scalar_t* __restrict__ value,  // [num_tokens, num_heads, head_size]
+    scalar_t* __restrict__ k_cache,      // [num_blocks, block_size, num_heads,
+                                         // head_size]
+    scalar_t* __restrict__ v_cache,      // [num_blocks, block_size, num_heads,
+                                         // head_size]
+    const int64_t* __restrict__ slot_mapping,  // [num_tokens]
+    const int block_stride, const int key_stride, const int value_stride,
+    const int num_heads, const int head_size, const int block_size) {
+  const int64_t token_idx = blockIdx.x;
+  const int64_t slot_idx = slot_mapping[token_idx];
+  // NOTE: slot_idx can be -1 if the token is padded
+  if (slot_idx < 0) {
+    return;
+  }
+  const int64_t block_idx = slot_idx / block_size;
+  const int64_t block_offset = slot_idx % block_size;
+  const int n = num_heads * head_size;
+  for (int i = threadIdx.x; i < n; i += blockDim.x) {
+    const int64_t src_key_idx = token_idx * key_stride + i;
+    const int64_t src_value_idx = token_idx * value_stride + i;
+    const int head_idx = i / head_size;
+    const int head_offset = i % head_size;
+    const int64_t tgt_value_idx = block_idx * block_stride +
+                                  block_offset * num_heads * head_size +
+                                  head_idx * head_size + head_offset;
+    k_cache[tgt_value_idx] = key[src_key_idx];
+    v_cache[tgt_value_idx] = value[src_value_idx];
+  }
+}
 }  // namespace vllm
 
-#define CALL_RESHAPE_AND_CACHE(KV_T, CACHE_T, IS_FP8_KV_CACHE)                                     \
+// KV_T is the stored data type of kv-cache.
-  vllm::reshape_and_cache_kernel<KV_T, CACHE_T, IS_FP8_KV_CACHE><<<grid, block, 0, stream>>>(      \
+// CACHE_T is the data type of key and value tensors.
+// KV_DTYPE is the real data type of kv-cache.
+#define CALL_RESHAPE_AND_CACHE(KV_T, CACHE_T, KV_DTYPE)               \
+  vllm::reshape_and_cache_kernel<KV_T, CACHE_T, KV_DTYPE>             \
+      <<<grid, block, 0, stream>>>(                                   \
           reinterpret_cast<KV_T*>(key.data_ptr()),                    \
           reinterpret_cast<KV_T*>(value.data_ptr()),                  \
           reinterpret_cast<CACHE_T*>(key_cache.data_ptr()),           \
           reinterpret_cast<CACHE_T*>(value_cache.data_ptr()),         \
-    slot_mapping.data_ptr<int64_t>(),                                                              \
+          slot_mapping.data_ptr<int64_t>(), key_stride, value_stride, \
-    key_stride,                                                                                    \
+          num_heads, head_size, block_size, x, kv_scale);
-    value_stride,                                                                                  \
-    num_heads,                                                                                     \
-    head_size,                                                                                     \
-    block_size,                                                                                    \
-    x,                                                                                             \
-    kv_scale);
 
 void reshape_and_cache(
     torch::Tensor& key,    // [num_tokens, num_heads, head_size]
     torch::Tensor& value,  // [num_tokens, num_heads, head_size]
-  torch::Tensor& key_cache,     // [num_blocks, num_heads, head_size/x, block_size, x]
+    torch::Tensor&
-  torch::Tensor& value_cache,   // [num_blocks, num_heads, head_size, block_size]
+        key_cache,  // [num_blocks, num_heads, head_size/x, block_size, x]
+    torch::Tensor&
+        value_cache,  // [num_blocks, num_heads, head_size, block_size]
     torch::Tensor& slot_mapping,  // [num_tokens]
-  const std::string& kv_cache_dtype,
+    const std::string& kv_cache_dtype, const double kv_scale) {
-  const float kv_scale)
-{
   int num_tokens = key.size(0);
   int num_heads = key.size(1);
   int head_size = key.size(2);
@@ -254,65 +272,77 @@ void reshape_and_cache(
   dim3 block(std::min(num_heads * head_size, 512));
   const at::cuda::OptionalCUDAGuard device_guard(device_of(key));
   const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
-  if (kv_cache_dtype == "auto") {
+
-    if (key.dtype() == at::ScalarType::Float) {
+  DISPATCH_BY_KV_CACHE_DTYPE(key.dtype(), kv_cache_dtype,
-      CALL_RESHAPE_AND_CACHE(float, float, false);
+                             CALL_RESHAPE_AND_CACHE)
-    } else if (key.dtype() == at::ScalarType::Half) {
-      CALL_RESHAPE_AND_CACHE(uint16_t, uint16_t, false);
-    } else if (key.dtype() == at::ScalarType::BFloat16) {
-      CALL_RESHAPE_AND_CACHE(__nv_bfloat16, __nv_bfloat16, false);
 }
-  } else if (kv_cache_dtype == "fp8") {
+
-    if (key.dtype() == at::ScalarType::Float) {
+void reshape_and_cache_flash(
-      CALL_RESHAPE_AND_CACHE(float, uint8_t, true);
+    torch::Tensor& key,      // [num_tokens, num_heads, head_size]
-    } else if (key.dtype() == at::ScalarType::Half) {
+    torch::Tensor& value,    // [num_tokens, num_heads, head_size]
-      CALL_RESHAPE_AND_CACHE(uint16_t, uint8_t, true);
+    torch::Tensor& k_cache,  // [num_blocks, block_size, num_heads, head_size]
-    } else if (key.dtype() == at::ScalarType::BFloat16) {
+    torch::Tensor& v_cache,  // [num_blocks, block_size, num_heads, head_size]
-      CALL_RESHAPE_AND_CACHE(__nv_bfloat16, uint8_t, true);
+    torch::Tensor& slot_mapping,  // [num_tokens]
-    }
+    const std::string& kv_cache_dtype) {
-  } else {
+  // FIXME: only support auto datatype, does not support fp8
+  if (kv_cache_dtype != "auto") {
     TORCH_CHECK(false, "Unsupported data type of kv cache: ", kv_cache_dtype);
   }
+  int num_tokens = key.size(0);
+  int num_heads = key.size(1);
+  int head_size = key.size(2);
+  int block_size = k_cache.size(1);
+
+  int key_stride = key.stride(0);
+  int value_stride = value.stride(0);
+  int block_stride = k_cache.stride(0);
+  TORCH_CHECK(k_cache.stride(0) == v_cache.stride(0));
+
+  dim3 grid(num_tokens);
+  dim3 block(std::min(num_heads * head_size, 512));
+  const at::cuda::OptionalCUDAGuard device_guard(device_of(key));
+  const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
+  VLLM_DISPATCH_FLOATING_TYPES(
+      key.scalar_type(), "reshape_and_cache_flash", [&] {
+        vllm::reshape_and_cache_flash_kernel<scalar_t>
+            <<<grid, block, 0, stream>>>(
+                key.data_ptr<scalar_t>(), value.data_ptr<scalar_t>(),
+                k_cache.data_ptr<scalar_t>(), v_cache.data_ptr<scalar_t>(),
+                slot_mapping.data_ptr<int64_t>(), block_stride, key_stride,
+                value_stride, num_heads, head_size, block_size);
+      });
 }
 
 namespace vllm {
 
-template<typename Tout, typename Tin>
+template <typename Tout, typename Tin, Fp8KVCacheDataType kv_dt>
-__global__ void convert_fp8_kernel(
+__global__ void convert_fp8_kernel(const Tin* __restrict__ src_cache,
-  const Tin* __restrict__ src_cache,
                                    Tout* __restrict__ dst_cache,
+                                   const float kv_scale,
                                    const int64_t block_stride) {
   const int64_t block_idx = blockIdx.x;
   for (int i = threadIdx.x; i < block_stride; i += blockDim.x) {
     int64_t idx = block_idx * block_stride + i;
-#if defined(ENABLE_FP8_E5M2)
+    dst_cache[idx] =
-    dst_cache[idx] = fp8_e5m2_unscaled::vec_conversion<Tout, Tin>(src_cache[idx]);
+        fp8::scaled_convert<Tout, Tin, kv_dt>(src_cache[idx], kv_scale);
-#elif defined(ENABLE_FP8_E4M3)
-    dst_cache[idx] = fp8_e4m3::vec_conversion<Tout, Tin>(src_cache[idx]);
-#else
-    assert(false);
-#endif
   }
 }
 
 }  // namespace vllm
 
-#define CALL_CONVERT_FP8(Tout, Tin)                                 \
+#define CALL_CONVERT_FP8(Tout, Tin, KV_DTYPE)                                \
-  vllm::convert_fp8_kernel<Tout, Tin><<<grid, block, 0, stream>>>(  \
+  vllm::convert_fp8_kernel<Tout, Tin, KV_DTYPE><<<grid, block, 0, stream>>>( \
       reinterpret_cast<Tin*>(src_cache.data_ptr()),                          \
-    reinterpret_cast<Tout*>(dst_cache.data_ptr()),                  \
+      reinterpret_cast<Tout*>(dst_cache.data_ptr()), kv_scale, block_stride);
-    block_stride);
 
-void convert_fp8(
+// Only for testing.
-  torch::Tensor& src_cache,
+void convert_fp8(torch::Tensor& dst_cache, torch::Tensor& src_cache,
-  torch::Tensor& dst_cache)
+                 const double kv_scale, const std::string& kv_cache_dtype) {
-{
   torch::Device src_device = src_cache.device();
   torch::Device dst_device = dst_cache.device();
   TORCH_CHECK(src_device.is_cuda(), "src must be on a GPU")
   TORCH_CHECK(dst_device.is_cuda(), "dst must be on a GPU")
-  TORCH_CHECK(
+  TORCH_CHECK(src_device.index() == dst_device.index(),
-    src_device.index() == dst_device.index(),
               "src and dst must be on the same GPU");
   at::cuda::OptionalCUDAGuard device_guard(src_device);
 
@@ -323,17 +353,37 @@ void convert_fp8(
   dim3 block(std::min(block_stride, int64_t(512)));
   const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
 
+  if (kv_cache_dtype == "auto") {
     if (src_cache.dtype() == at::ScalarType::Float) {
-    CALL_CONVERT_FP8(uint8_t, float);
+      CALL_CONVERT_FP8(uint8_t, float, vllm::Fp8KVCacheDataType::kAuto);
     } else if (src_cache.dtype() == at::ScalarType::Half) {
-    CALL_CONVERT_FP8(uint8_t, uint16_t);
+      CALL_CONVERT_FP8(uint8_t, uint16_t, vllm::Fp8KVCacheDataType::kAuto);
     } else if (src_cache.dtype() == at::ScalarType::BFloat16) {
-    CALL_CONVERT_FP8(uint8_t, __nv_bfloat16);
+      CALL_CONVERT_FP8(uint8_t, __nv_bfloat16, vllm::Fp8KVCacheDataType::kAuto);
     } else if (dst_cache.dtype() == at::ScalarType::Float) {
-    CALL_CONVERT_FP8(float, uint8_t);
+      CALL_CONVERT_FP8(float, uint8_t, vllm::Fp8KVCacheDataType::kAuto);
     } else if (dst_cache.dtype() == at::ScalarType::Half) {
-    CALL_CONVERT_FP8(uint16_t, uint8_t);
+      CALL_CONVERT_FP8(uint16_t, uint8_t, vllm::Fp8KVCacheDataType::kAuto);
     } else if (dst_cache.dtype() == at::ScalarType::BFloat16) {
-    CALL_CONVERT_FP8(__nv_bfloat16, uint8_t);
+      CALL_CONVERT_FP8(__nv_bfloat16, uint8_t, vllm::Fp8KVCacheDataType::kAuto);
+    }
+  } else if (kv_cache_dtype == "fp8" || kv_cache_dtype == "fp8_e4m3") {
+    if (src_cache.dtype() == at::ScalarType::Float) {
+      CALL_CONVERT_FP8(uint8_t, float, vllm::Fp8KVCacheDataType::kFp8E4M3);
+    } else if (src_cache.dtype() == at::ScalarType::Half) {
+      CALL_CONVERT_FP8(uint8_t, uint16_t, vllm::Fp8KVCacheDataType::kFp8E4M3);
+    } else if (src_cache.dtype() == at::ScalarType::BFloat16) {
+      CALL_CONVERT_FP8(uint8_t, __nv_bfloat16,
+                       vllm::Fp8KVCacheDataType::kFp8E4M3);
+    } else if (dst_cache.dtype() == at::ScalarType::Float) {
+      CALL_CONVERT_FP8(float, uint8_t, vllm::Fp8KVCacheDataType::kFp8E4M3);
+    } else if (dst_cache.dtype() == at::ScalarType::Half) {
+      CALL_CONVERT_FP8(uint16_t, uint8_t, vllm::Fp8KVCacheDataType::kFp8E4M3);
+    } else if (dst_cache.dtype() == at::ScalarType::BFloat16) {
+      CALL_CONVERT_FP8(__nv_bfloat16, uint8_t,
+                       vllm::Fp8KVCacheDataType::kFp8E4M3);
+    }
+  } else {
+    TORCH_CHECK(false, "Unsupported data type: ", kv_cache_dtype);
   }
 }

csrc/cpu/activation.cpp

@@ -81,12 +81,10 @@ void silu_and_mul(torch::Tensor &out, torch::Tensor &input) {
   int num_tokens = input.numel() / input.size(-1);
   int d = input.size(-1) / 2;
 
-  VLLM_DISPATCH_FLOATING_TYPES(
+  VLLM_DISPATCH_FLOATING_TYPES(input.scalar_type(), "silu_and_mul_impl", [&] {
-      input.scalar_type(), "silu_and_mul_impl", [&] {
     CPU_KERNEL_GUARD_IN(silu_and_mul_impl)
-        activation_kernel<scalar_t, silu_act, true>(num_tokens, d,
+    activation_kernel<scalar_t, silu_act, true>(
-                                                    input.data_ptr<scalar_t>(),
+        num_tokens, d, input.data_ptr<scalar_t>(), out.data_ptr<scalar_t>());
-                                                    out.data_ptr<scalar_t>());
     CPU_KERNEL_GUARD_OUT(silu_and_mul_impl)
   });
 }
@@ -97,12 +95,10 @@ void gelu_and_mul(torch::Tensor &out,   // [..., d]
   int num_tokens = input.numel() / input.size(-1);
   int d = input.size(-1) / 2;
 
-  VLLM_DISPATCH_FLOATING_TYPES(
+  VLLM_DISPATCH_FLOATING_TYPES(input.scalar_type(), "gelu_and_mul_impl", [&] {
-      input.scalar_type(), "gelu_and_mul_impl", [&] {
     CPU_KERNEL_GUARD_IN(gelu_and_mul_impl)
-        activation_kernel<scalar_t, gelu_act, true>(num_tokens, d,
+    activation_kernel<scalar_t, gelu_act, true>(
-                                                    input.data_ptr<scalar_t>(),
+        num_tokens, d, input.data_ptr<scalar_t>(), out.data_ptr<scalar_t>());
-                                                    out.data_ptr<scalar_t>());
     CPU_KERNEL_GUARD_OUT(gelu_and_mul_impl)
   });
 }

csrc/cpu/attention.cpp

@@ -2,7 +2,8 @@
 
 namespace {
 
-template <typename scalar_t> struct KernelVecType {
+template <typename scalar_t>
+struct KernelVecType {
   using q_load_vec_type = void;
   using q_vec_type = void;
   using k_load_vec_type = void;
@@ -11,7 +12,8 @@ template <typename scalar_t> struct KernelVecType {
   using v_load_vec_type = void;
 };
 
-template <> struct KernelVecType<float> {
+template <>
+struct KernelVecType<float> {
   using q_load_vec_type = vec_op::FP32Vec4;
   using q_vec_type = vec_op::FP32Vec16;
   using k_load_vec_type = vec_op::FP32Vec16;
@@ -21,7 +23,8 @@ template <> struct KernelVecType<float> {
 };
 
 #ifdef __AVX512BF16__
-template <> struct KernelVecType<c10::BFloat16> {
+template <>
+struct KernelVecType<c10::BFloat16> {
   using q_load_vec_type = vec_op::BF16Vec8;
   using q_vec_type = vec_op::BF16Vec32;
   using k_load_vec_type = vec_op::BF16Vec32;
@@ -30,7 +33,8 @@ template <> struct KernelVecType<c10::BFloat16> {
   using v_load_vec_type = vec_op::BF16Vec16;
 };
 #else
-template <> struct KernelVecType<c10::BFloat16> {
+template <>
+struct KernelVecType<c10::BFloat16> {
   using q_load_vec_type = vec_op::BF16Vec8;
   using q_vec_type = vec_op::FP32Vec16;
   using k_load_vec_type = vec_op::BF16Vec16;
@@ -67,14 +71,15 @@ FORCE_INLINE std::pair<T, T> reduceSoftmax(T *data, const int size,
 }
 
 template <typename T>
-FORCE_INLINE std::pair<T, T>
+FORCE_INLINE std::pair<T, T> reduceSoftmaxAlibi(T* data, const int size,
-reduceSoftmaxAlibi(T *data, const int size, const int capacity,
+                                                const int capacity,
-                   const float alibi_slope, const int start_index,
+                                                const float alibi_slope,
-                   const int context_len) {
+                                                const int start_index,
-  data[0] += alibi_slope * (start_index - context_len + 1);
+                                                const int seq_len) {
+  data[0] += alibi_slope * (start_index - seq_len + 1);
   T max = data[0];
   for (int i = 1; i < size; ++i) {
-    T qk = data[i] + alibi_slope * (start_index + i - context_len + 1);
+    T qk = data[i] + alibi_slope * (start_index + i - seq_len + 1);
     data[i] = qk;
     max = max >= qk ? max : qk;
   }
@@ -215,17 +220,17 @@ FORCE_INLINE void reduceValueBlock(const float *prob, const scalar_t *v_block,
 namespace {
 template <typename scalar_t, int HEAD_SIZE, int BLOCK_SIZE>
 struct paged_attention_v1_impl {
-  static void
+  static void call(
-  call(scalar_t *__restrict__ out,           // [num_seqs, num_heads, head_size]
+      scalar_t* __restrict__ out,            // [num_seqs, num_heads, head_size]
       const scalar_t* __restrict__ q,        // [num_seqs, num_heads, head_size]
       const scalar_t* __restrict__ k_cache,  // [num_blocks, num_kv_heads,
                                              // head_size/x, block_size, x]
       const scalar_t* __restrict__ v_cache,  // [num_blocks, num_kv_heads,
                                              // head_size, block_size]
       const int num_kv_heads, const float scale,
-       const int
+      const int* __restrict__ block_tables,  // [num_seqs,
-           *__restrict__ block_tables, // [num_seqs, max_num_blocks_per_seq]
+                                             // max_num_blocks_per_seq]
-       const int *__restrict__ context_lens, // [num_seqs]
+      const int* __restrict__ seq_lens,      // [num_seqs]
       const int max_num_blocks_per_seq,
       const float* __restrict__ alibi_slopes,  // [num_heads]
       const int q_stride, const int kv_block_stride, const int kv_head_stride,
@@ -235,32 +240,31 @@ struct paged_attention_v1_impl {
 
     static_assert(BLOCK_SIZE == 16);
 
-    int max_context_len = max_num_blocks_per_seq * BLOCK_SIZE;
+    int max_seq_len = max_num_blocks_per_seq * BLOCK_SIZE;
-    int max_context_len_padded = (max_context_len + 15) & 0xFFFFFFF0;
+    int max_seq_len_padded = (max_seq_len + 15) & 0xFFFFFFF0;
-    TORCH_CHECK((max_context_len_padded * sizeof(float)) % 64 == 0);
+    TORCH_CHECK((max_seq_len_padded * sizeof(float)) % 64 == 0);
 
     const int parallel_work_item_num = omp_get_max_threads();
 
     size_t logits_bytes =
-        parallel_work_item_num * max_context_len_padded * sizeof(float);
+        parallel_work_item_num * max_seq_len_padded * sizeof(float);
     float* logits = (float*)std::aligned_alloc(
         64, logits_bytes);  // Cacheline alignment for each context token.
-                           // [parallel_work_item_num, max_context_len_padded]
+                            // [parallel_work_item_num, max_seq_len_padded]
 
 #pragma omp parallel for collapse(2) schedule(dynamic, 1)
     for (int seq_idx = 0; seq_idx < num_seqs; ++seq_idx) {
       for (int head_idx = 0; head_idx < num_heads; ++head_idx) {
-        int context_len = context_lens[seq_idx];
+        int seq_len = seq_lens[seq_idx];
         const int* seq_block_table =
             block_tables + max_num_blocks_per_seq * seq_idx;
-        const int block_num = (context_len + BLOCK_SIZE - 1) / BLOCK_SIZE;
+        const int block_num = (seq_len + BLOCK_SIZE - 1) / BLOCK_SIZE;
         const int64_t kv_head_idx = head_idx / num_queries_per_kv;
         const scalar_t* __restrict__ q_vec_ptr =
             q + seq_idx * q_stride + head_idx * HEAD_SIZE;
-        const int last_block_token_num =
+        const int last_block_token_num = seq_len - (block_num - 1) * BLOCK_SIZE;
-            context_len - (block_num - 1) * BLOCK_SIZE;
         float* __restrict__ thread_block_logits =
-            logits + omp_get_thread_num() * max_context_len_padded;
+            logits + omp_get_thread_num() * max_seq_len_padded;
 
         // Compute logits
         for (int block_idx = 0; block_idx < block_num; ++block_idx) {
@@ -278,12 +282,11 @@ struct paged_attention_v1_impl {
 
         // Compute softmax
         if (alibi_slopes) {
-          reduceSoftmaxAlibi(thread_block_logits, context_len,
+          reduceSoftmaxAlibi(thread_block_logits, seq_len,
                              block_num * BLOCK_SIZE, alibi_slopes[head_idx], 0,
-                             context_len);
+                             seq_len);
         } else {
-          reduceSoftmax(thread_block_logits, context_len,
+          reduceSoftmax(thread_block_logits, seq_len, block_num * BLOCK_SIZE);
-                        block_num * BLOCK_SIZE);
         }
 
         // Compute value
@@ -340,7 +343,7 @@ struct paged_attention_v1_impl {
 #define LAUNCH_V1_ATTENTION_KERNEL(T, HEAD_SIZE, BLOCK_SIZE)                   \
   paged_attention_v1_impl<T, HEAD_SIZE, BLOCK_SIZE>::call(                     \
       out_ptr, query_ptr, key_cache_ptr, value_cache_ptr, num_kv_heads, scale, \
-      block_tables_ptr, context_lens_ptr, max_num_blocks_per_seq,              \
+      block_tables_ptr, seq_lens_ptr, max_num_blocks_per_seq,                  \
       alibi_slopes_ptr, q_stride, kv_block_stride, kv_head_stride, num_seqs,   \
       num_heads);
 
@@ -348,8 +351,8 @@ template <typename T, int BLOCK_SIZE>
 void paged_attention_v1_impl_launcher(
     torch::Tensor& out, torch::Tensor& query, torch::Tensor& key_cache,
     torch::Tensor& value_cache, int num_kv_heads, float scale,
-    torch::Tensor &block_tables, torch::Tensor &context_lens,
+    torch::Tensor& block_tables, torch::Tensor& seq_lens, int max_seq_len,
-    int max_context_len, const c10::optional<torch::Tensor> &alibi_slopes) {
+    const c10::optional<torch::Tensor>& alibi_slopes) {
   int num_seqs = query.size(0);
   int num_heads = query.size(1);
   int head_size = query.size(2);
@@ -369,7 +372,7 @@ void paged_attention_v1_impl_launcher(
   T* key_cache_ptr = reinterpret_cast<T*>(key_cache.data_ptr());
   T* value_cache_ptr = reinterpret_cast<T*>(value_cache.data_ptr());
   int* block_tables_ptr = block_tables.data_ptr<int>();
-  int *context_lens_ptr = context_lens.data_ptr<int>();
+  int* seq_lens_ptr = seq_lens.data_ptr<int>();
 
   switch (head_size) {
     case 64:
@@ -387,6 +390,9 @@ void paged_attention_v1_impl_launcher(
     case 128:
       LAUNCH_V1_ATTENTION_KERNEL(T, 128, BLOCK_SIZE);
       break;
+    case 192:
+      LAUNCH_V1_ATTENTION_KERNEL(T, 192, BLOCK_SIZE);
+      break;
     case 256:
       LAUNCH_V1_ATTENTION_KERNEL(T, 256, BLOCK_SIZE);
       break;
@@ -399,7 +405,7 @@ void paged_attention_v1_impl_launcher(
 #define CALL_V1_KERNEL_LAUNCHER(T, BLOCK_SIZE)                               \
   paged_attention_v1_impl_launcher<T, BLOCK_SIZE>(                           \
       out, query, key_cache, value_cache, num_kv_heads, scale, block_tables, \
-      context_lens, max_context_len, alibi_slopes);
+      seq_lens, max_seq_len, alibi_slopes);
 
 #define CALL_V1_KERNEL_LAUNCHER_BLOCK_SIZE(T)                     \
   switch (block_size) {                                           \
@@ -412,15 +418,18 @@ void paged_attention_v1_impl_launcher(
   }
 }  // namespace
 
-void paged_attention_v1(torch::Tensor &out, torch::Tensor &query,
+void paged_attention_v1(
-                        torch::Tensor &key_cache, torch::Tensor &value_cache,
+    torch::Tensor& out, torch::Tensor& query, torch::Tensor& key_cache,
-                        int num_kv_heads, float scale,
+    torch::Tensor& value_cache, int64_t num_kv_heads, double scale,
-                        torch::Tensor &block_tables,
+    torch::Tensor& block_tables, torch::Tensor& seq_lens, int64_t block_size,
-                        torch::Tensor &context_lens, int block_size,
+    int64_t max_seq_len, const c10::optional<torch::Tensor>& alibi_slopes,
-                        int max_context_len,
+    const std::string& kv_cache_dtype, double kv_scale, const int64_t tp_rank,
-                        const c10::optional<torch::Tensor> &alibi_slopes,
+    const int64_t blocksparse_local_blocks,
-                        const std::string &kv_cache_dtype, float kv_scale) {
+    const int64_t blocksparse_vert_stride, const int64_t blocksparse_block_size,
+    const int64_t blocksparse_head_sliding_step) {
   TORCH_CHECK(kv_scale == 1.0f);
+  TORCH_CHECK(blocksparse_vert_stride <= 1,
+              "CPU backend does not support blocksparse attention yet.");
   VLLM_DISPATCH_FLOATING_TYPES(query.scalar_type(), "paged_attention_v1_impl",
                                [&] {
                                  CPU_KERNEL_GUARD_IN(paged_attention_v1_impl)
@@ -435,9 +444,10 @@ template <typename scalar_t, int HEAD_SIZE, int BLOCK_SIZE, int PARTITION_SIZE>
 struct paged_attention_v2_impl {
   static void call(
       scalar_t* __restrict__ out,            // [num_seqs, num_heads, head_size]
-      float *__restrict__ exp_sums, // [num_seqs, num_heads, max_num_partitions]
+      float* __restrict__ exp_sums,          // [num_seqs, num_heads,
-      float
+                                             // max_num_partitions]
-          *__restrict__ max_logits, // [num_seqs, num_heads, max_num_partitions]
+      float* __restrict__ max_logits,        // [num_seqs, num_heads,
+                                             // max_num_partitions]
       scalar_t* __restrict__ tmp_out,        // [num_seqs, num_heads,
                                              // max_num_partitions, head_size]
       const scalar_t* __restrict__ q,        // [num_seqs, num_heads, head_size]
@@ -446,9 +456,9 @@ struct paged_attention_v2_impl {
       const scalar_t* __restrict__ v_cache,  // [num_blocks, num_kv_heads,
                                              // head_size, block_size]
       const int num_kv_heads, const float scale,
-      const int
+      const int* __restrict__ block_tables,  // [num_seqs,
-          *__restrict__ block_tables, // [num_seqs, max_num_blocks_per_seq]
+                                             // max_num_blocks_per_seq]
-      const int *__restrict__ context_lens, // [num_seqs]
+      const int* __restrict__ seq_lens,      // [num_seqs]
       const int max_num_blocks_per_seq,
       const float* __restrict__ alibi_slopes,  // [num_heads]
       const int q_stride, const int kv_block_stride, const int kv_head_stride,
@@ -465,22 +475,20 @@ struct paged_attention_v2_impl {
       for (int partition_idx = 0; partition_idx < max_num_partitions;
            ++partition_idx) {
         for (int head_idx = 0; head_idx < num_heads; ++head_idx) {
-          const int context_len = context_lens[seq_idx];
+          const int seq_len = seq_lens[seq_idx];
           const int start_token_idx = partition_idx * PARTITION_SIZE;
 
-          if (start_token_idx >= context_len)
+          if (start_token_idx >= seq_len) continue;
-            continue;
 
           const int partition_num =
-              (context_len + PARTITION_SIZE - 1) / PARTITION_SIZE;
+              (seq_len + PARTITION_SIZE - 1) / PARTITION_SIZE;
           const bool no_reduce = (partition_num == 1);
-          const int context_token_num =
+          const int token_num =
-              (std::min(context_len, start_token_idx + PARTITION_SIZE) -
+              (std::min(seq_len, start_token_idx + PARTITION_SIZE) -
                start_token_idx);
-          const int block_num =
+          const int block_num = (token_num + BLOCK_SIZE - 1) / BLOCK_SIZE;
-              (context_token_num + BLOCK_SIZE - 1) / BLOCK_SIZE;
           const int last_block_token_num =
-              context_token_num - (block_num - 1) * BLOCK_SIZE;
+              token_num - (block_num - 1) * BLOCK_SIZE;
           const int* seq_block_table = block_tables +
                                        max_num_blocks_per_seq * seq_idx +
                                        start_token_idx / BLOCK_SIZE;
@@ -507,11 +515,11 @@ struct paged_attention_v2_impl {
           std::pair<float, float> max_and_sum;
           if (alibi_slopes) {
             max_and_sum = reduceSoftmaxAlibi(
-                logits, context_token_num, block_num * BLOCK_SIZE,
+                logits, token_num, block_num * BLOCK_SIZE,
-                alibi_slopes[head_idx], start_token_idx, context_len);
+                alibi_slopes[head_idx], start_token_idx, seq_len);
           } else {
-            max_and_sum = reduceSoftmax(logits, context_token_num,
+            max_and_sum =
-                                        block_num * BLOCK_SIZE);
+                reduceSoftmax(logits, token_num, block_num * BLOCK_SIZE);
           }
 
           auto&& [max_logit, exp_sum] = max_and_sum;
@@ -583,12 +591,11 @@ struct paged_attention_v2_impl {
 #pragma omp parallel for collapse(2) schedule(static, 1)
     for (int seq_idx = 0; seq_idx < num_seqs; ++seq_idx) {
       for (int head_idx = 0; head_idx < num_heads; ++head_idx) {
-        const int context_len = context_lens[seq_idx];
+        const int seq_len = seq_lens[seq_idx];
         const int partition_num =
-            (context_len + PARTITION_SIZE - 1) / PARTITION_SIZE;
+            (seq_len + PARTITION_SIZE - 1) / PARTITION_SIZE;
 
-        if (partition_num == 1)
+        if (partition_num == 1) continue;
-          continue;
 
         reducePartitonSoftmax(
             max_logits + seq_idx * num_heads * max_num_partitions +
@@ -603,8 +610,8 @@ struct paged_attention_v2_impl {
     using v_load_vec_type = typename KernelVecType<scalar_t>::v_load_vec_type;
     static_assert(v_load_vec_type::get_elem_num() == BLOCK_SIZE);
     constexpr int head_elem_num_per_group =
-        16; // Note: didn't align with the cacheline size, due to some HEAD_SIZE
+        16;  // Note: didn't align with the cacheline size, due to some
-            // didn't align with 64 bytes
+             // HEAD_SIZE didn't align with 64 bytes
     static_assert(HEAD_SIZE % head_elem_num_per_group == 0);
     constexpr int head_group_num = HEAD_SIZE / head_elem_num_per_group;
     const float* __restrict__ rescale_factors = exp_sums;
@@ -612,12 +619,11 @@ struct paged_attention_v2_impl {
     for (int seq_idx = 0; seq_idx < num_seqs; ++seq_idx) {
       for (int head_idx = 0; head_idx < num_heads; ++head_idx) {
         for (int group_idx = 0; group_idx < head_group_num; ++group_idx) {
-          const int context_len = context_lens[seq_idx];
+          const int seq_len = seq_lens[seq_idx];
           const int partition_num =
-              (context_len + PARTITION_SIZE - 1) / PARTITION_SIZE;
+              (seq_len + PARTITION_SIZE - 1) / PARTITION_SIZE;
 
-          if (partition_num == 1)
+          if (partition_num == 1) continue;
-            continue;
 
           const float* __restrict__ seq_head_rescale_factors =
               rescale_factors + seq_idx * num_heads * max_num_partitions +
@@ -649,7 +655,7 @@ struct paged_attention_v2_impl {
   paged_attention_v2_impl<T, HEAD_SIZE, BLOCK_SIZE, PARTITION_SIZE>::call(   \
       out_ptr, exp_sums_ptr, max_logits_ptr, tmp_out_ptr, query_ptr,         \
       key_cache_ptr, value_cache_ptr, num_kv_heads, scale, block_tables_ptr, \
-      context_lens_ptr, max_num_blocks_per_seq, alibi_slopes_ptr, q_stride,    \
+      seq_lens_ptr, max_num_blocks_per_seq, alibi_slopes_ptr, q_stride,      \
       kv_block_stride, kv_head_stride, num_seqs, num_heads,                  \
       max_num_partitions);
 
@@ -658,8 +664,8 @@ void paged_attention_v2_impl_launcher(
     torch::Tensor& out, torch::Tensor& exp_sums, torch::Tensor& max_logits,
     torch::Tensor& tmp_out, torch::Tensor& query, torch::Tensor& key_cache,
     torch::Tensor& value_cache, int num_kv_heads, float scale,
-    torch::Tensor &block_tables, torch::Tensor &context_lens, int block_size,
+    torch::Tensor& block_tables, torch::Tensor& seq_lens, int block_size,
-    int max_context_len, const c10::optional<torch::Tensor> &alibi_slopes) {
+    int max_seq_len, const c10::optional<torch::Tensor>& alibi_slopes) {
   int num_seqs = query.size(0);
   int num_heads = query.size(1);
   int head_size = query.size(2);
@@ -683,7 +689,7 @@ void paged_attention_v2_impl_launcher(
   T* key_cache_ptr = reinterpret_cast<T*>(key_cache.data_ptr());
   T* value_cache_ptr = reinterpret_cast<T*>(value_cache.data_ptr());
   int* block_tables_ptr = block_tables.data_ptr<int>();
-  int *context_lens_ptr = context_lens.data_ptr<int>();
+  int* seq_lens_ptr = seq_lens.data_ptr<int>();
 
   switch (head_size) {
     case 64:
@@ -701,6 +707,9 @@ void paged_attention_v2_impl_launcher(
     case 128:
       LAUNCH_V2_ATTENTION_KERNEL(T, 128, BLOCK_SIZE);
       break;
+    case 192:
+      LAUNCH_V2_ATTENTION_KERNEL(T, 192, BLOCK_SIZE);
+      break;
     case 256:
       LAUNCH_V2_ATTENTION_KERNEL(T, 256, BLOCK_SIZE);
       break;
@@ -713,8 +722,8 @@ void paged_attention_v2_impl_launcher(
 #define CALL_V2_KERNEL_LAUNCHER(T, BLOCK_SIZE)                              \
   paged_attention_v2_impl_launcher<T, BLOCK_SIZE>(                          \
       out, exp_sums, max_logits, tmp_out, query, key_cache, value_cache,    \
-      num_kv_heads, scale, block_tables, context_lens, block_size,             \
+      num_kv_heads, scale, block_tables, seq_lens, block_size, max_seq_len, \
-      max_context_len, alibi_slopes);
+      alibi_slopes);
 
 #define CALL_V2_KERNEL_LAUNCHER_BLOCK_SIZE(T)                     \
   switch (block_size) {                                           \
@@ -727,16 +736,19 @@ void paged_attention_v2_impl_launcher(
   }
 }  // namespace
 
-void paged_attention_v2(torch::Tensor &out, torch::Tensor &exp_sums,
+void paged_attention_v2(
-                        torch::Tensor &max_logits, torch::Tensor &tmp_out,
+    torch::Tensor& out, torch::Tensor& exp_sums, torch::Tensor& max_logits,
-                        torch::Tensor &query, torch::Tensor &key_cache,
+    torch::Tensor& tmp_out, torch::Tensor& query, torch::Tensor& key_cache,
-                        torch::Tensor &value_cache, int num_kv_heads,
+    torch::Tensor& value_cache, int64_t num_kv_heads, double scale,
-                        float scale, torch::Tensor &block_tables,
+    torch::Tensor& block_tables, torch::Tensor& seq_lens, int64_t block_size,
-                        torch::Tensor &context_lens, int block_size,
+    int64_t max_seq_len, const c10::optional<torch::Tensor>& alibi_slopes,
-                        int max_context_len,
+    const std::string& kv_cache_dtype, double kv_scale, const int64_t tp_rank,
-                        const c10::optional<torch::Tensor> &alibi_slopes,
+    const int64_t blocksparse_local_blocks,
-                        const std::string &kv_cache_dtype, float kv_scale) {
+    const int64_t blocksparse_vert_stride, const int64_t blocksparse_block_size,
+    const int64_t blocksparse_head_sliding_step) {
   TORCH_CHECK(kv_scale == 1.0f);
+  TORCH_CHECK(blocksparse_vert_stride <= 1,
+              "CPU backend does not support blocksparse attention yet.");
   VLLM_DISPATCH_FLOATING_TYPES(query.scalar_type(), "paged_attention_v2_impl",
                                [&] {
                                  CPU_KERNEL_GUARD_IN(paged_attention_v2_impl)

csrc/cpu/cache.cpp

@@ -5,19 +5,20 @@
 
 namespace {
 template <typename scalar_t>
-void copy_blocks_cpu_impl(
+void copy_blocks_cpu_impl(std::vector<torch::Tensor> const& key_caches,
-    std::vector<torch::Tensor> &key_caches,
+                          std::vector<torch::Tensor> const& value_caches,
-    std::vector<torch::Tensor> &value_caches,
+                          const torch::Tensor& mapping_pairs,
-    const std::vector<std::pair<int64_t, int64_t>> mapping_pairs,
+                          const int element_num_per_block,
-    const int element_num_per_block, const int layer_num) {
+                          const int layer_num) {
-  const size_t pair_num = mapping_pairs.size();
+  const size_t pair_num = mapping_pairs.size(0);
   const size_t block_bytes = sizeof(scalar_t) * element_num_per_block;
 #pragma omp parallel for collapse(2)
   for (int layer = 0; layer < layer_num; ++layer) {
     for (size_t pair = 0; pair < pair_num; ++pair) {
-      int64_t source_offset = element_num_per_block * mapping_pairs[pair].first;
+      int64_t source_offset =
+          element_num_per_block * mapping_pairs[pair][0].item<int64_t>();
       int64_t target_offset =
-          element_num_per_block * mapping_pairs[pair].second;
+          element_num_per_block * mapping_pairs[pair][1].item<int64_t>();
       scalar_t* key_cache_ptr = key_caches[layer].data_ptr<scalar_t>();
       scalar_t* source_ptr = key_cache_ptr + source_offset;
       scalar_t* target_ptr = key_cache_ptr + target_offset;
@@ -81,28 +82,23 @@ void reshape_and_cache_cpu_impl(
 }
 };  // namespace
 
-void copy_blocks(std::vector<torch::Tensor> &key_caches,
+// Note: the key_caches and value_caches vectors are constant but
-                 std::vector<torch::Tensor> &value_caches,
+// not the Tensors they contain. The vectors need to be const refs
-                 const std::map<int64_t, std::vector<int64_t>> &block_mapping) {
+// in order to satisfy pytorch's C++ operator registration code.
-  int num_layers = key_caches.size();
+void copy_blocks(std::vector<torch::Tensor> const& key_caches,
+                 std::vector<torch::Tensor> const& value_caches,
+                 const torch::Tensor& block_mapping) {
+  unsigned num_layers = key_caches.size();
   TORCH_CHECK(num_layers == value_caches.size());
   if (num_layers == 0) {
     return;
   }
 
-  std::vector<std::pair<int64_t, int64_t>> mapping_pairs;
-  mapping_pairs.reserve(block_mapping.size());
-  for (const auto &pair : block_mapping) {
-    for (const auto &dst : pair.second) {
-      mapping_pairs.emplace_back(pair.first, dst);
-    }
-  }
-
   const int element_num_per_block = key_caches[0][0].numel();
   VLLM_DISPATCH_FLOATING_TYPES(
       key_caches[0].scalar_type(), "copy_blocks_cpu_impl", [&] {
         CPU_KERNEL_GUARD_IN(copy_blocks_cpu_impl)
-        copy_blocks_cpu_impl<scalar_t>(key_caches, value_caches, mapping_pairs,
+        copy_blocks_cpu_impl<scalar_t>(key_caches, value_caches, block_mapping,
                                        element_num_per_block, num_layers);
         CPU_KERNEL_GUARD_OUT(copy_blocks_cpu_impl)
       });
@@ -111,7 +107,7 @@ void copy_blocks(std::vector<torch::Tensor> &key_caches,
 void reshape_and_cache(torch::Tensor& key, torch::Tensor& value,
                        torch::Tensor& key_cache, torch::Tensor& value_cache,
                        torch::Tensor& slot_mapping,
-                       const std::string &kv_cache_dtype, float kv_scale) {
+                       const std::string& kv_cache_dtype, double kv_scale) {
   TORCH_CHECK(kv_scale == 1.0f);
 
   int num_tokens = key.size(0);
@@ -136,6 +132,6 @@ void reshape_and_cache(torch::Tensor &key, torch::Tensor &value,
 }
 
 void swap_blocks(torch::Tensor& src, torch::Tensor& dst,
-                 const std::map<int64_t, int64_t> &block_mapping) {
+                 const torch::Tensor& block_mapping) {
   TORCH_CHECK(false, "swap_blocks is unsupported on CPU.")
 }

csrc/cpu/cpu_types.hpp

@@ -3,7 +3,7 @@
 #define CPU_TYPES_HPP
 
 #include <immintrin.h>
-#include <torch/extension.h>
+#include <torch/all.h>
 
 namespace vec_op {
 

csrc/cpu/layernorm.cpp

@@ -87,8 +87,8 @@ void fused_add_rms_norm_impl(scalar_t *__restrict__ input,
 }
 }  // namespace
 
-void rms_norm(torch::Tensor &out, torch::Tensor &input,
+void rms_norm(torch::Tensor& out, torch::Tensor& input, torch::Tensor& weight,
-                  torch::Tensor &weight, float epsilon) {
+              double epsilon) {
   int hidden_size = input.size(-1);
   int num_tokens = input.numel() / hidden_size;
 
@@ -102,7 +102,7 @@ void rms_norm(torch::Tensor &out, torch::Tensor &input,
 }
 
 void fused_add_rms_norm(torch::Tensor& input, torch::Tensor& residual,
-                            torch::Tensor &weight, float epsilon) {
+                        torch::Tensor& weight, double epsilon) {
   int hidden_size = input.size(-1);
   int num_tokens = input.numel() / hidden_size;
 

csrc/cpu/pos_encoding.cpp

@@ -4,25 +4,74 @@
 namespace {
 template <typename scalar_t>
 void rotary_embedding_impl(
-    const int64_t
+    const int64_t* __restrict__ positions,  // [batch_size, seq_len] or
-        *__restrict__ positions, // [batch_size, seq_len] or [num_tokens]
+                                            // [num_tokens]
-    scalar_t
+    scalar_t* __restrict__ query,           /// [batch_size, seq_len, num_heads,
-        *__restrict__ query, /// [batch_size, seq_len, num_heads, head_size] or
+                                   /// head_size] or [num_tokens, num_heads,
-                             /// [num_tokens, num_heads, head_size]
+                                   /// head_size]
-    scalar_t
+    scalar_t* __restrict__ key,  // [batch_size, seq_len, num_kv_heads,
-        *__restrict__ key, // [batch_size, seq_len, num_kv_heads, head_size] or
+                                 // head_size] or [num_tokens, num_kv_heads,
-                           // [num_tokens, num_kv_heads, head_size]
+                                 // head_size]
-    const scalar_t
+    const scalar_t* __restrict__ cos_sin_cache,  // [max_position, 2, rot_dim //
-        *__restrict__ cos_sin_cache, // [max_position, 2, rot_dim // 2]
+                                                 // 2]
     const int rot_dim, const int64_t query_stride, const int64_t key_stride,
     const int num_heads, const int num_kv_heads, const int head_size,
     const int num_tokens) {
   using scalar_vec_t = vec_op::vec_t<scalar_t>;
   constexpr int VEC_ELEM_NUM = scalar_vec_t::get_elem_num();
-  constexpr int ELEM_SIZE = sizeof(scalar_t);
 
   const int embed_dim = rot_dim / 2;
-  TORCH_CHECK(embed_dim % VEC_ELEM_NUM == 0);
+  bool flag = (embed_dim % VEC_ELEM_NUM == 0);
+  const int loop_upper = flag ? embed_dim : embed_dim - VEC_ELEM_NUM;
+
+  auto compute_loop = [&](const int64_t token_head, const scalar_t* cache_ptr,
+                          scalar_t* qk) {
+    int j = 0;
+    for (; j < loop_upper; j += VEC_ELEM_NUM) {
+      const int rot_offset = j;
+      const int x_index = rot_offset;
+      const int y_index = embed_dim + rot_offset;
+
+      const int64_t out_x = token_head + x_index;
+      const int64_t out_y = token_head + y_index;
+
+      const scalar_vec_t cos(cache_ptr + x_index);
+      const scalar_vec_t sin(cache_ptr + y_index);
+
+      const scalar_vec_t q_x(qk + out_x);
+      const scalar_vec_t q_y(qk + out_y);
+
+      vec_op::FP32Vec8 fp32_cos(cos);
+      vec_op::FP32Vec8 fp32_sin(sin);
+
+      vec_op::FP32Vec8 fp32_q_x(q_x);
+      vec_op::FP32Vec8 fp32_q_y(q_y);
+
+      auto out1 = fp32_q_x * fp32_cos - fp32_q_y * fp32_sin;
+      scalar_vec_t(out1).save(qk + out_x);
+
+      auto out2 = fp32_q_y * fp32_cos + fp32_q_x * fp32_sin;
+      scalar_vec_t(out2).save(qk + out_y);
+    }
+    if (!flag) {
+      for (; j < embed_dim; ++j) {
+        const int x_index = j;
+        const int y_index = embed_dim + j;
+
+        const int64_t out_x = token_head + x_index;
+        const int64_t out_y = token_head + y_index;
+
+        const float fp32_cos = cache_ptr[x_index];
+        const float fp32_sin = cache_ptr[y_index];
+
+        const float fp32_q_x = qk[out_x];
+        const float fp32_q_y = qk[out_y];
+
+        qk[out_x] = fp32_q_x * fp32_cos - fp32_q_y * fp32_sin;
+        qk[out_y] = fp32_q_y * fp32_cos + fp32_q_x * fp32_sin;
+      }
+    }
+  };
 
 #pragma omp parallel for
   for (int token_idx = 0; token_idx < num_tokens; ++token_idx) {
@@ -33,78 +82,29 @@ void rotary_embedding_impl(
       const int head_idx = i;
       const int64_t token_head =
           token_idx * query_stride + head_idx * head_size;
-      for (int j = 0; j < embed_dim; j += VEC_ELEM_NUM) {
+      compute_loop(token_head, cache_ptr, query);
-        const int rot_offset = j;
-        const int x_index = rot_offset;
-        const int y_index = embed_dim + rot_offset;
-
-        const int64_t out_x = token_head + x_index;
-        const int64_t out_y = token_head + y_index;
-
-        const scalar_vec_t cos(cache_ptr + x_index);
-        const scalar_vec_t sin(cache_ptr + y_index);
-
-        const scalar_vec_t q_x(query + out_x);
-        const scalar_vec_t q_y(query + out_y);
-
-        vec_op::FP32Vec8 fp32_cos(cos);
-        vec_op::FP32Vec8 fp32_sin(sin);
-
-        vec_op::FP32Vec8 fp32_q_x(q_x);
-        vec_op::FP32Vec8 fp32_q_y(q_y);
-
-        auto out1 = fp32_q_x * fp32_cos - fp32_q_y * fp32_sin;
-        scalar_vec_t(out1).save(query + out_x);
-
-        auto out2 = fp32_q_y * fp32_cos + fp32_q_x * fp32_sin;
-        scalar_vec_t(out2).save(query + out_y);
-      }
     }
 
     for (int i = 0; i < num_kv_heads; ++i) {
       const int head_idx = i;
       const int64_t token_head = token_idx * key_stride + head_idx * head_size;
-      for (int j = 0; j < embed_dim; j += VEC_ELEM_NUM) {
+      compute_loop(token_head, cache_ptr, key);
-        const int rot_offset = j;
-        const int x_index = rot_offset;
-        const int y_index = embed_dim + rot_offset;
-
-        const int64_t out_x = token_head + x_index;
-        const int64_t out_y = token_head + y_index;
-
-        const scalar_vec_t cos(cache_ptr + x_index);
-        const scalar_vec_t sin(cache_ptr + y_index);
-
-        const scalar_vec_t k_x(key + out_x);
-        const scalar_vec_t k_y(key + out_y);
-
-        vec_op::FP32Vec8 fp32_cos(cos);
-        vec_op::FP32Vec8 fp32_sin(sin);
-
-        vec_op::FP32Vec8 fp32_k_x(k_x);
-        vec_op::FP32Vec8 fp32_k_y(k_y);
-
-        auto out1 = fp32_k_x * fp32_cos - fp32_k_y * fp32_sin;
-        scalar_vec_t(out1).save(key + out_x);
-        auto out2 = fp32_k_y * fp32_cos + fp32_k_x * fp32_sin;
-        scalar_vec_t(out2).save(key + out_y);
-      }
     }
   }
 }
 
 template <typename scalar_t>
 void rotary_embedding_gptj_impl(
-    const int64_t
+    const int64_t* __restrict__ positions,  // [batch_size, seq_len] or
-        *__restrict__ positions, // [batch_size, seq_len] or [num_tokens]
+                                            // [num_tokens]
-    scalar_t
+    scalar_t* __restrict__ query,           /// [batch_size, seq_len, num_heads,
-        *__restrict__ query, /// [batch_size, seq_len, num_heads, head_size] or
+                                   /// head_size] or [num_tokens, num_heads,
-                             /// [num_tokens, num_heads, head_size]
+                                   /// head_size]
-    scalar_t
+    scalar_t* __restrict__ key,  // [batch_size, seq_len, num_kv_heads,
-        *__restrict__ key, // [batch_size, seq_len, num_kv_heads, head_size] or
+                                 // head_size] or [num_tokens, num_kv_heads,
-                           // [num_tokens, num_kv_heads, head_size]
+                                 // head_size]
-    const scalar_t
+    const scalar_t* __restrict__ cos_sin_cache,  // [max_position, 2, rot_dim //
-        *__restrict__ cos_sin_cache, // [max_position, 2, rot_dim // 2]
+                                                 // 2]
     const int rot_dim, const int64_t query_stride, const int64_t key_stride,
     const int num_heads, const int num_kv_heads, const int head_size,
     const int num_tokens) {
@@ -168,7 +168,7 @@ void rotary_embedding_gptj_impl(
 };  // namespace
 
 void rotary_embedding(torch::Tensor& positions, torch::Tensor& query,
-                          torch::Tensor &key, int head_size,
+                      torch::Tensor& key, int64_t head_size,
                       torch::Tensor& cos_sin_cache, bool is_neox) {
   int num_tokens = query.numel() / query.size(-1);
   int rot_dim = cos_sin_cache.size(1);

csrc/cpu/pybind.cpp

@@ -1,73 +0,0 @@
-#include "cache.h"
-#include "cuda_utils.h"
-#include "ops.h"
-#include <torch/extension.h>
-
-PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
-  // vLLM custom ops
-  pybind11::module ops = m.def_submodule("ops", "vLLM custom operators");
-
-  // Attention ops
-  ops.def(
-    "paged_attention_v1",
-    &paged_attention_v1,
-    "Compute the attention between an input query and the cached keys/values using PagedAttention.");
-  ops.def(
-    "paged_attention_v2",
-    &paged_attention_v2,
-    "PagedAttention V2.");
-
-  // Activation ops
-  ops.def(
-    "silu_and_mul",
-    &silu_and_mul,
-    "Activation function used in SwiGLU.");
-  ops.def(
-    "gelu_and_mul",
-    &gelu_and_mul,
-    "Activation function used in GeGLU with `none` approximation.");
-  ops.def(
-    "gelu_tanh_and_mul",
-    &gelu_tanh_and_mul,
-    "Activation function used in GeGLU with `tanh` approximation.");
-  ops.def(
-    "gelu_new",
-    &gelu_new,
-    "GELU implementation used in GPT-2.");
-  ops.def(
-    "gelu_fast",
-    &gelu_fast,
-    "Approximate GELU implementation.");
-
-  // Layernorm
-  ops.def(
-    "rms_norm",
-    &rms_norm,
-    "Apply Root Mean Square (RMS) Normalization to the input tensor.");
-
-  ops.def(
-    "fused_add_rms_norm",
-    &fused_add_rms_norm,
-    "In-place fused Add and RMS Normalization");
-
-  // Rotary embedding
-  ops.def(
-    "rotary_embedding",
-    &rotary_embedding,
-    "Apply GPT-NeoX or GPT-J style rotary embedding to query and key");
-
-  // Cache ops
-  pybind11::module cache_ops = m.def_submodule("cache_ops", "vLLM cache ops");
-  cache_ops.def(
-    "swap_blocks",
-    &swap_blocks,
-    "Swap in (out) the cache blocks from src to dst");
-  cache_ops.def(
-    "copy_blocks",
-    &copy_blocks,
-    "Copy the cache blocks from src to dst");
-  cache_ops.def(
-    "reshape_and_cache",
-    &reshape_and_cache,
-    "Reshape the key and value tensors and cache them");
-}

csrc/cpu/torch_bindings.cpp

@@ -0,0 +1,106 @@
+#include "cache.h"
+#include "ops.h"
+#include "registration.h"
+
+#include <torch/library.h>
+
+TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, ops) {
+  // vLLM custom ops
+
+  // Attention ops
+  // Compute the attention between an input query and the cached keys/values
+  // using PagedAttention.
+  ops.def(
+      "paged_attention_v1("
+      "    Tensor! out, Tensor query, Tensor key_cache,"
+      "    Tensor value_cache, int num_kv_heads, float scale,"
+      "    Tensor block_tables, Tensor seq_lens, int block_size,"
+      "    int max_seq_len, Tensor? alibi_slopes,"
+      "    str kv_cache_dtype, float kv_scale, int tp_rank,"
+      "    int blocksparse_local_blocks,"
+      "    int blocksparse_vert_stride, int blocksparse_block_size,"
+      "    int blocksparse_head_sliding_step) -> ()");
+  ops.impl("paged_attention_v1", torch::kCPU, &paged_attention_v1);
+
+  // PagedAttention V2.
+  ops.def(
+      "paged_attention_v2("
+      "    Tensor! out, Tensor exp_sums, Tensor max_logits,"
+      "    Tensor tmp_out, Tensor query, Tensor key_cache,"
+      "    Tensor value_cache, int num_kv_heads, float scale,"
+      "    Tensor block_tables, Tensor seq_lens, int block_size,"
+      "    int max_seq_len, Tensor? alibi_slopes,"
+      "    str kv_cache_dtype, float kv_scale, int tp_rank,"
+      "    int blocksparse_local_blocks,"
+      "    int blocksparse_vert_stride, int blocksparse_block_size,"
+      "    int blocksparse_head_sliding_step) -> ()");
+  ops.impl("paged_attention_v2", torch::kCPU, &paged_attention_v2);
+
+  // Activation ops
+
+  // Activation function used in SwiGLU.
+  ops.def("silu_and_mul(Tensor! out, Tensor input) -> ()");
+  ops.impl("silu_and_mul", torch::kCPU, &silu_and_mul);
+
+  // Activation function used in GeGLU with `none` approximation.
+  ops.def("gelu_and_mul(Tensor! out, Tensor input) -> ()");
+  ops.impl("gelu_and_mul", torch::kCPU, &gelu_and_mul);
+
+  // Activation function used in GeGLU with `tanh` approximation.
+  ops.def("gelu_tanh_and_mul(Tensor! out, Tensor input) -> ()");
+  ops.impl("gelu_tanh_and_mul", torch::kCPU, &gelu_tanh_and_mul);
+
+  // GELU implementation used in GPT-2.
+  ops.def("gelu_new(Tensor! out, Tensor input) -> ()");
+  ops.impl("gelu_new", torch::kCPU, &gelu_new);
+
+  // Approximate GELU implementation.
+  ops.def("gelu_fast(Tensor! out, Tensor input) -> ()");
+  ops.impl("gelu_fast", torch::kCPU, &gelu_fast);
+
+  // Layernorm
+  // Apply Root Mean Square (RMS) Normalization to the input tensor.
+  ops.def(
+      "rms_norm(Tensor! out, Tensor input, Tensor weight, float epsilon) -> "
+      "()");
+  ops.impl("rms_norm", torch::kCPU, &rms_norm);
+
+  // In-place fused Add and RMS Normalization.
+  ops.def(
+      "fused_add_rms_norm(Tensor! input, Tensor! residual, Tensor weight, "
+      "float epsilon) -> ()");
+  ops.impl("fused_add_rms_norm", torch::kCPU, &fused_add_rms_norm);
+
+  // Rotary embedding
+  // Apply GPT-NeoX or GPT-J style rotary embedding to query and key.
+  ops.def(
+      "rotary_embedding(Tensor positions, Tensor! query,"
+      "                 Tensor! key, int head_size,"
+      "                 Tensor cos_sin_cache, bool is_neox) -> ()");
+  ops.impl("rotary_embedding", torch::kCPU, &rotary_embedding);
+}
+
+TORCH_LIBRARY_EXPAND(CONCAT(TORCH_EXTENSION_NAME, _cache_ops), cache_ops) {
+  // Cache ops
+  // Swap in (out) the cache blocks from src to dst.
+  cache_ops.def(
+      "swap_blocks(Tensor src, Tensor! dst, Tensor block_mapping) -> ()");
+  cache_ops.impl("swap_blocks", torch::kCPU, &swap_blocks);
+
+  // Copy the cache blocks from src to dst.
+  cache_ops.def(
+      "copy_blocks(Tensor[]! key_caches, Tensor[]! value_caches, Tensor "
+      "block_mapping) -> ()");
+  cache_ops.impl("copy_blocks", torch::kCPU, &copy_blocks);
+
+  // Reshape the key and value tensors and cache them.
+  cache_ops.def(
+      "reshape_and_cache(Tensor key, Tensor value,"
+      "                  Tensor! key_cache, Tensor! value_cache,"
+      "                  Tensor slot_mapping,"
+      "                  str kv_cache_dtype,"
+      "                  float kv_scale) -> ()");
+  cache_ops.impl("reshape_and_cache", torch::kCPU, &reshape_and_cache);
+}
+
+REGISTER_EXTENSION(TORCH_EXTENSION_NAME)

csrc/cuda_compat.h

@@ -17,9 +17,14 @@
 #endif
 
 #ifndef USE_ROCM
-  #define VLLM_SHFL_XOR_SYNC(var, lane_mask) __shfl_xor_sync(uint32_t(-1), var, lane_mask)
+  #define VLLM_SHFL_XOR_SYNC(var, lane_mask) \
+    __shfl_xor_sync(uint32_t(-1), var, lane_mask)
+  #define VLLM_SHFL_XOR_SYNC_WIDTH(var, lane_mask, width) \
+    __shfl_xor_sync(uint32_t(-1), var, lane_mask, width)
 #else
   #define VLLM_SHFL_XOR_SYNC(var, lane_mask) __shfl_xor(var, lane_mask)
+  #define VLLM_SHFL_XOR_SYNC_WIDTH(var, lane_mask, width) \
+    __shfl_xor(var, lane_mask, width)
 #endif
 
 #ifndef USE_ROCM
@@ -28,6 +33,13 @@
   #define VLLM_SHFL_SYNC(var, src_lane) __shfl(var, src_lane)
 #endif
 
+#ifndef USE_ROCM
+  #define VLLM_SHFL_DOWN_SYNC(var, lane_delta) \
+    __shfl_down_sync(uint32_t(-1), var, lane_delta)
+#else
+  #define VLLM_SHFL_DOWN_SYNC(var, lane_delta) __shfl_down(var, lane_delta)
+#endif
+
 #ifndef USE_ROCM
   #define VLLM_DevFuncAttribute_SET_MaxDynamicSharedMemorySize(FUNC, VAL) \
     cudaFuncSetAttribute(FUNC, cudaFuncAttributeMaxDynamicSharedMemorySize, VAL)
@@ -35,4 +47,3 @@
   #define VLLM_DevFuncAttribute_SET_MaxDynamicSharedMemorySize(FUNC, VAL) \
     hipFuncSetAttribute(FUNC, hipFuncAttributeMaxDynamicSharedMemorySize, VAL)
 #endif
-

csrc/cuda_utils.h

@@ -1,10 +1,5 @@
 #pragma once
 
-#include <torch/extension.h>
+int64_t get_device_attribute(int64_t attribute, int64_t device_id);
 
-int get_device_attribute(
+int64_t get_max_shared_memory_per_block_device_attribute(int64_t device_id);
-    int attribute,
-    int device_id);
-
-int get_max_shared_memory_per_block_device_attribute(
-    int device_id);

csrc/cuda_utils_kernels.cu

@@ -2,26 +2,20 @@
   #include <hip/hip_runtime.h>
   #include <hip/hip_runtime_api.h>
 #endif
-int get_device_attribute(
+int64_t get_device_attribute(int64_t attribute, int64_t device_id) {
-    int attribute,
-    int device_id)
-{
   int device, value;
   if (device_id < 0) {
     cudaGetDevice(&device);
-    }
+  } else {
-    else {
     device = device_id;
   }
-    cudaDeviceGetAttribute(&value, static_cast<cudaDeviceAttr>(attribute), device);
+  cudaDeviceGetAttribute(&value, static_cast<cudaDeviceAttr>(attribute),
+                         device);
   return value;
 }
 
-
+int64_t get_max_shared_memory_per_block_device_attribute(int64_t device_id) {
-int get_max_shared_memory_per_block_device_attribute(
+  int64_t attribute;
-    int device_id)
-{
-int attribute;    
   // https://docs.nvidia.com/cuda/cuda-runtime-api/group__CUDART__TYPES.html
   // cudaDevAttrMaxSharedMemoryPerBlockOptin = 97 if not is_hip() else 74
 

csrc/custom_all_reduce.cu

@@ -1,17 +1,17 @@
 #include <ATen/cuda/Exceptions.h>
 #include <c10/cuda/CUDAGuard.h>
 #include <c10/cuda/CUDAStream.h>
-#include <torch/extension.h>
+#include <torch/all.h>
 
 #include "custom_all_reduce.cuh"
 
-// fake pointer type
+// fake pointer type, must match fptr_t type in ops.h
-using fptr_t = uint64_t;
+using fptr_t = int64_t;
 static_assert(sizeof(void*) == sizeof(fptr_t));
 
 fptr_t init_custom_ar(torch::Tensor& meta, torch::Tensor& rank_data,
                       const std::vector<std::string>& handles,
-                      const std::vector<int64_t> &offsets, int rank,
+                      const std::vector<int64_t>& offsets, int64_t rank,
                       bool full_nvlink) {
   int world_size = offsets.size();
   if (world_size > 8)
@@ -55,7 +55,7 @@ bool _is_weak_contiguous(torch::Tensor &t) {
           t.numel() * t.element_size());
 }
 
-bool should_custom_ar(torch::Tensor &inp, int max_size, int world_size,
+bool should_custom_ar(torch::Tensor& inp, int64_t max_size, int64_t world_size,
                       bool full_nvlink) {
   auto inp_size = inp.numel() * inp.element_size();
   // custom allreduce requires input byte size to be multiples of 16
@@ -80,8 +80,7 @@ void _all_reduce(fptr_t _fa, torch::Tensor &inp, torch::Tensor &out,
     }
     case at::ScalarType::Half: {
       fa->allreduce<half>(stream, reinterpret_cast<half*>(inp.data_ptr()),
-                          reinterpret_cast<half *>(out.data_ptr()),
+                          reinterpret_cast<half*>(out.data_ptr()), out.numel());
-                          out.numel());
       break;
     }
 #if (__CUDA_ARCH__ >= 800 || !defined(__CUDA_ARCH__))
@@ -126,7 +125,7 @@ void dispose(fptr_t _fa) {
   delete fa;
 }
 
-int meta_size() { return sizeof(vllm::Signal); }
+int64_t meta_size() { return sizeof(vllm::Signal); }
 
 void register_buffer(fptr_t _fa, torch::Tensor& t,
                      const std::vector<std::string>& handles,
@@ -135,10 +134,16 @@ void register_buffer(fptr_t _fa, torch::Tensor &t,
   fa->register_buffer(handles, offsets, t.data_ptr());
 }
 
-std::pair<std::vector<uint8_t>, std::vector<int64_t>> get_graph_buffer_ipc_meta(
+std::tuple<torch::Tensor, std::vector<int64_t>> get_graph_buffer_ipc_meta(
     fptr_t _fa) {
   auto fa = reinterpret_cast<vllm::CustomAllreduce*>(_fa);
-  return fa->get_graph_buffer_ipc_meta();
+  auto [handle_bytes, offsets] = fa->get_graph_buffer_ipc_meta();
+  auto options =
+      torch::TensorOptions().dtype(torch::kUInt8).device(torch::kCPU);
+  auto handles =
+      torch::empty({static_cast<int64_t>(handle_bytes.size())}, options);
+  std::memcpy(handles.data_ptr(), handle_bytes.data(), handle_bytes.size());
+  return {handles, std::move(offsets)};
 }
 
 void register_graph_buffers(fptr_t _fa, const std::vector<std::string>& handles,

csrc/custom_all_reduce.cuh

@@ -137,8 +137,7 @@ DINLINE void start_sync(const RankSignals &sg, volatile Signal *self_sg,
     // Latency = 1 p2p write
     sg.signals[threadIdx.x]->start[blockIdx.x][rank] = 1;
     // wait until we got true from all ranks
-    while (!self_sg->start[blockIdx.x][threadIdx.x])
+    while (!self_sg->start[blockIdx.x][threadIdx.x]);
-      ;
   }
   __syncthreads();
 }
@@ -162,8 +161,7 @@ DINLINE void end_sync(const RankSignals &sg, volatile Signal *self_sg,
     // Latency = 1 p2p write
     sg.signals[threadIdx.x]->end[blockIdx.x][rank] = 1;
     // wait until we got true from all ranks
-    while (!self_sg->end[blockIdx.x][threadIdx.x])
+    while (!self_sg->end[blockIdx.x][threadIdx.x]);
-      ;
   }
   if constexpr (!final_sync) __syncthreads();
 }
@@ -192,8 +190,7 @@ __global__ void __launch_bounds__(512, 1)
   // do the actual reduction
   for (int idx = blockIdx.x * blockDim.x + threadIdx.x; idx < size;
        idx += gridDim.x * blockDim.x) {
-    ((P *)result)[idx] =
+    ((P*)result)[idx] = packed_reduce<P, ngpus, A>((const P**)&dp.ptrs[0], idx);
-        packed_reduce<P, ngpus, A>((const P **)&dp.ptrs[0], idx);
   }
   end_sync<ngpus, true>(sg, self_sg, rank);
 }

csrc/dispatch_utils.h

@@ -4,7 +4,7 @@
  */
 #pragma once
 
-#include <torch/extension.h>
+#include <torch/all.h>
 
 #define VLLM_DISPATCH_CASE_FLOATING_TYPES(...)         \
   AT_DISPATCH_CASE(at::ScalarType::Float, __VA_ARGS__) \
@@ -12,8 +12,7 @@
   AT_DISPATCH_CASE(at::ScalarType::BFloat16, __VA_ARGS__)
 
 #define VLLM_DISPATCH_FLOATING_TYPES(TYPE, NAME, ...) \
-  AT_DISPATCH_SWITCH(                                             \
+  AT_DISPATCH_SWITCH(TYPE, NAME, VLLM_DISPATCH_CASE_FLOATING_TYPES(__VA_ARGS__))
-    TYPE, NAME, VLLM_DISPATCH_CASE_FLOATING_TYPES(__VA_ARGS__))
 
 #define VLLM_DISPATCH_CASE_FLOATING_AND_BYTE_TYPES(...)   \
   AT_DISPATCH_CASE(at::ScalarType::Float, __VA_ARGS__)    \
@@ -22,8 +21,8 @@
   AT_DISPATCH_CASE(at::ScalarType::Byte, __VA_ARGS__)
 
 #define VLLM_DISPATCH_FLOATING_AND_BYTE_TYPES(TYPE, NAME, ...) \
-  AT_DISPATCH_SWITCH(                                                    \
+  AT_DISPATCH_SWITCH(TYPE, NAME,                               \
-    TYPE, NAME, VLLM_DISPATCH_CASE_FLOATING_AND_BYTE_TYPES(__VA_ARGS__))
+                     VLLM_DISPATCH_CASE_FLOATING_AND_BYTE_TYPES(__VA_ARGS__))
 
 #define VLLM_DISPATCH_CASE_INTEGRAL_TYPES(...)         \
   AT_DISPATCH_CASE(at::ScalarType::Byte, __VA_ARGS__)  \
@@ -33,5 +32,4 @@
   AT_DISPATCH_CASE(at::ScalarType::Long, __VA_ARGS__)
 
 #define VLLM_DISPATCH_INTEGRAL_TYPES(TYPE, NAME, ...) \
-  AT_DISPATCH_SWITCH(                                             \
+  AT_DISPATCH_SWITCH(TYPE, NAME, VLLM_DISPATCH_CASE_INTEGRAL_TYPES(__VA_ARGS__))
-    TYPE, NAME, VLLM_DISPATCH_CASE_INTEGRAL_TYPES(__VA_ARGS__))

csrc/layernorm_kernels.cu

@@ -1,4 +1,4 @@
-#include <torch/extension.h>
+#include <torch/all.h>
 #include <ATen/cuda/CUDAContext.h>
 #include <c10/cuda/CUDAGuard.h>
 
@@ -23,9 +23,7 @@ __global__ void rms_norm_kernel(
     scalar_t* __restrict__ out,           // [..., hidden_size]
     const scalar_t* __restrict__ input,   // [..., hidden_size]
     const scalar_t* __restrict__ weight,  // [hidden_size]
-  const float epsilon,
+    const float epsilon, const int num_tokens, const int hidden_size) {
-  const int num_tokens,
-  const int hidden_size) {
   __shared__ float s_variance;
   float variance = 0.0f;
 
@@ -41,11 +39,11 @@ __global__ void rms_norm_kernel(
 
   for (int idx = threadIdx.x; idx < hidden_size; idx += blockDim.x) {
     float x = (float)input[blockIdx.x * hidden_size + idx];
-    out[blockIdx.x * hidden_size + idx] = ((scalar_t) (x * s_variance)) * weight[idx];
+    out[blockIdx.x * hidden_size + idx] =
+        ((scalar_t)(x * s_variance)) * weight[idx];
   }
 }
 
-
 /* Converter structs for the conversion from torch types to HIP/CUDA types,
    and the associated type conversions within HIP/CUDA. These helpers need
    to be implemented for now because the relevant type conversion
@@ -57,7 +55,9 @@ __global__ void rms_norm_kernel(
    If true, the struct should be fully defined as shown in the examples below.
  */
 template <typename torch_type>
-struct _typeConvert { static constexpr bool exists = false; };
+struct _typeConvert {
+  static constexpr bool exists = false;
+};
 
 #if defined(USE_ROCM) || (defined(CUDA_VERSION) && (CUDA_VERSION >= 12000))
 // CUDA < 12.0 runs into issues with packed type conversion
@@ -68,9 +68,15 @@ struct _typeConvert<c10::Half> {
   using packed_hip_type = __half2;
 
   __device__ static inline float convert(hip_type x) { return __half2float(x); }
-  __device__ static inline float2 convert(packed_hip_type x) { return __half22float2(x); }
+  __device__ static inline float2 convert(packed_hip_type x) {
-  __device__ static inline hip_type convert(float x) { return __float2half_rn(x); }
+    return __half22float2(x);
-  __device__ static inline packed_hip_type convert(float2 x) { return __float22half2_rn(x); }
+  }
+  __device__ static inline hip_type convert(float x) {
+    return __float2half_rn(x);
+  }
+  __device__ static inline packed_hip_type convert(float2 x) {
+    return __float22half2_rn(x);
+  }
 };
 
   #if defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 800
@@ -82,13 +88,22 @@ struct _typeConvert<c10::BFloat16> {
   using hip_type = __nv_bfloat16;
   using packed_hip_type = __nv_bfloat162;
 
-  __device__ static inline float convert(hip_type x) { return __bfloat162float(x); }
+  __device__ static inline float convert(hip_type x) {
-  __device__ static inline float2 convert(packed_hip_type x) { return __bfloat1622float2(x); }
+    return __bfloat162float(x);
-  __device__ static inline hip_type convert(float x) { return __float2bfloat16(x); }
+  }
-  __device__ static inline packed_hip_type convert(float2 x) { return __float22bfloat162_rn(x); }
+  __device__ static inline float2 convert(packed_hip_type x) {
+    return __bfloat1622float2(x);
+  }
+  __device__ static inline hip_type convert(float x) {
+    return __float2bfloat16(x);
+  }
+  __device__ static inline packed_hip_type convert(float2 x) {
+    return __float22bfloat162_rn(x);
+  }
 };
   #endif  // defined(__CUDA_ARCH__) && __CUDA_ARCH__ >= 800
-#endif // defined(USE_ROCM) || (defined(CUDA_VERSION) && (CUDA_VERSION >= 12000))
+#endif    // defined(USE_ROCM) || (defined(CUDA_VERSION) && (CUDA_VERSION >=
+          // 12000))
 
 /* Vector POD struct to generate vectorized and packed FP16/BF16 ops
    for appropriate specializations of fused_add_rms_norm_kernel.
@@ -117,8 +132,7 @@ struct alignas(16) _f16Vec {
       }
     } else {
 #pragma unroll
-      for (int i = 0; i < width; ++i)
+      for (int i = 0; i < width; ++i) data[i] += other.data[i];
-        data[i] += other.data[i];
     }
     return *this;
   }
@@ -134,8 +148,7 @@ struct alignas(16) _f16Vec {
       }
     } else {
 #pragma unroll
-      for (int i = 0; i < width; ++i)
+      for (int i = 0; i < width; ++i) data[i] *= other.data[i];
-        data[i] *= other.data[i];
     }
     return *this;
   }
@@ -185,14 +198,12 @@ struct alignas(16) _f16Vec {
    packed and vectorized operations, which help with the
    memory latency bottleneck. */
 template <typename scalar_t, int width>
-__global__ std::enable_if_t<
+__global__ std::enable_if_t<(width > 0) && _typeConvert<scalar_t>::exists>
-  (width > 0) && _typeConvert<scalar_t>::exists> fused_add_rms_norm_kernel(
+fused_add_rms_norm_kernel(
     scalar_t* __restrict__ input,         // [..., hidden_size]
     scalar_t* __restrict__ residual,      // [..., hidden_size]
     const scalar_t* __restrict__ weight,  // [hidden_size]
-  const float epsilon,
+    const float epsilon, const int num_tokens, const int hidden_size) {
-  const int num_tokens,
-  const int hidden_size) {
   // Sanity checks on our vector struct and type-punned pointer arithmetic
   static_assert(std::is_pod_v<_f16Vec<scalar_t, width>>);
   static_assert(sizeof(_f16Vec<scalar_t, width>) == sizeof(scalar_t) * width);
@@ -203,9 +214,12 @@ __global__ std::enable_if_t<
   /* These and the argument pointers are all declared `restrict` as they are
      not aliased in practice. Argument pointers should not be dereferenced
      in this kernel as that would be undefined behavior */
-  auto* __restrict__ input_v = reinterpret_cast<_f16Vec<scalar_t, width>*>(input);
+  auto* __restrict__ input_v =
-  auto* __restrict__ residual_v = reinterpret_cast<_f16Vec<scalar_t, width>*>(residual);
+      reinterpret_cast<_f16Vec<scalar_t, width>*>(input);
-  auto* __restrict__ weight_v = reinterpret_cast<const _f16Vec<scalar_t, width>*>(weight);
+  auto* __restrict__ residual_v =
+      reinterpret_cast<_f16Vec<scalar_t, width>*>(residual);
+  auto* __restrict__ weight_v =
+      reinterpret_cast<const _f16Vec<scalar_t, width>*>(weight);
 
   for (int idx = threadIdx.x; idx < vec_hidden_size; idx += blockDim.x) {
     int id = blockIdx.x * vec_hidden_size + idx;
@@ -218,7 +232,8 @@ __global__ std::enable_if_t<
      calculation of max_block_size in fused_add_rms_norm */
   if (num_tokens < 256) {
     variance = blockReduceSum<float, 1024>(variance);
-  } else variance = blockReduceSum<float, 256>(variance);
+  } else
+    variance = blockReduceSum<float, 256>(variance);
   if (threadIdx.x == 0) {
     s_variance = rsqrtf(variance / hidden_size + epsilon);
   }
@@ -233,19 +248,16 @@ __global__ std::enable_if_t<
   }
 }
 
-
 /* Generic fused_add_rms_norm_kernel
    The width field is not used here but necessary for other specializations.
  */
 template <typename scalar_t, int width>
-__global__ std::enable_if_t<
+__global__ std::enable_if_t<(width == 0) || !_typeConvert<scalar_t>::exists>
-  (width == 0) || !_typeConvert<scalar_t>::exists> fused_add_rms_norm_kernel(
+fused_add_rms_norm_kernel(
     scalar_t* __restrict__ input,         // [..., hidden_size]
     scalar_t* __restrict__ residual,      // [..., hidden_size]
     const scalar_t* __restrict__ weight,  // [hidden_size]
-  const float epsilon,
+    const float epsilon, const int num_tokens, const int hidden_size) {
-  const int num_tokens,
-  const int hidden_size) {
   __shared__ float s_variance;
   float variance = 0.0f;
 
@@ -260,7 +272,8 @@ __global__ std::enable_if_t<
      calculation of max_block_size in fused_add_rms_norm */
   if (num_tokens < 256) {
     variance = blockReduceSum<float, 1024>(variance);
-  } else variance = blockReduceSum<float, 256>(variance);
+  } else
+    variance = blockReduceSum<float, 256>(variance);
   if (threadIdx.x == 0) {
     s_variance = rsqrtf(variance / hidden_size + epsilon);
   }
@@ -268,17 +281,17 @@ __global__ std::enable_if_t<
 
   for (int idx = threadIdx.x; idx < hidden_size; idx += blockDim.x) {
     float x = (float)residual[blockIdx.x * hidden_size + idx];
-    input[blockIdx.x * hidden_size + idx] = ((scalar_t) (x * s_variance)) * weight[idx];
+    input[blockIdx.x * hidden_size + idx] =
+        ((scalar_t)(x * s_variance)) * weight[idx];
   }
 }
 
 }  // namespace vllm
 
-void rms_norm(
+void rms_norm(torch::Tensor& out,     // [..., hidden_size]
-  torch::Tensor& out,      // [..., hidden_size]
               torch::Tensor& input,   // [..., hidden_size]
               torch::Tensor& weight,  // [hidden_size]
-  float epsilon) {
+              double epsilon) {
   int hidden_size = input.size(-1);
   int num_tokens = input.numel() / hidden_size;
 
@@ -286,40 +299,27 @@ void rms_norm(
   dim3 block(std::min(hidden_size, 1024));
   const at::cuda::OptionalCUDAGuard device_guard(device_of(input));
   const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
-  VLLM_DISPATCH_FLOATING_TYPES(
+  VLLM_DISPATCH_FLOATING_TYPES(input.scalar_type(), "rms_norm_kernel", [&] {
-    input.scalar_type(),
-    "rms_norm_kernel",
-    [&] {
     vllm::rms_norm_kernel<scalar_t><<<grid, block, 0, stream>>>(
-        out.data_ptr<scalar_t>(),
+        out.data_ptr<scalar_t>(), input.data_ptr<scalar_t>(),
-        input.data_ptr<scalar_t>(),
+        weight.data_ptr<scalar_t>(), epsilon, num_tokens, hidden_size);
-        weight.data_ptr<scalar_t>(),
-        epsilon,
-        num_tokens,
-        hidden_size);
   });
 }
 
 #define LAUNCH_FUSED_ADD_RMS_NORM(width)                                       \
   VLLM_DISPATCH_FLOATING_TYPES(                                                \
-    input.scalar_type(),                              \
+      input.scalar_type(), "fused_add_rms_norm_kernel", [&] {                  \
-    "fused_add_rms_norm_kernel",                      \
+        vllm::fused_add_rms_norm_kernel<scalar_t, width>                       \
-    [&] {                                             \
+            <<<grid, block, 0, stream>>>(input.data_ptr<scalar_t>(),           \
-      vllm::fused_add_rms_norm_kernel                 \
-      <scalar_t, width><<<grid, block, 0, stream>>>(  \
-        input.data_ptr<scalar_t>(),                   \
                                          residual.data_ptr<scalar_t>(),        \
-        weight.data_ptr<scalar_t>(),                  \
+                                         weight.data_ptr<scalar_t>(), epsilon, \
-        epsilon,                                      \
+                                         num_tokens, hidden_size);             \
-        num_tokens,                                   \
-        hidden_size);                                 \
       });
 
-void fused_add_rms_norm(
+void fused_add_rms_norm(torch::Tensor& input,     // [..., hidden_size]
-  torch::Tensor& input,    // [..., hidden_size]
                         torch::Tensor& residual,  // [..., hidden_size]
                         torch::Tensor& weight,    // [hidden_size]
-  float epsilon) {
+                        double epsilon) {
   int hidden_size = input.size(-1);
   int num_tokens = input.numel() / hidden_size;
 
@@ -342,8 +342,8 @@ void fused_add_rms_norm(
   auto inp_ptr = reinterpret_cast<std::uintptr_t>(input.data_ptr());
   auto res_ptr = reinterpret_cast<std::uintptr_t>(residual.data_ptr());
   auto wt_ptr = reinterpret_cast<std::uintptr_t>(weight.data_ptr());
-  bool ptrs_are_aligned = inp_ptr % 16 == 0 && res_ptr % 16 == 0 \
+  bool ptrs_are_aligned =
-                          && wt_ptr % 16 == 0;
+      inp_ptr % 16 == 0 && res_ptr % 16 == 0 && wt_ptr % 16 == 0;
   if (ptrs_are_aligned && hidden_size % 8 == 0) {
     LAUNCH_FUSED_ADD_RMS_NORM(8);
   } else {

csrc/moe/moe_ops.cpp

@@ -1,7 +0,0 @@
-#include "moe_ops.h"
-
-#include <torch/extension.h>
-
-PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
-  m.def("topk_softmax", &topk_softmax, "Apply topk softmax to the gating outputs.");
-}

csrc/moe/moe_ops.h

@@ -1,9 +1,7 @@
 #pragma once
 
-#include <torch/extension.h>
+#include <torch/all.h>
 
-void topk_softmax(
+void topk_softmax(torch::Tensor& topk_weights, torch::Tensor& topk_indices,
-  torch::Tensor& topk_weights,
-  torch::Tensor& topk_indices,
                   torch::Tensor& token_expert_indices,
                   torch::Tensor& gating_output);

csrc/moe/topk_softmax_kernels.cu

@@ -16,18 +16,25 @@
  * See the License for the specific language governing permissions and
  * limitations under the License.
  */
-#include <torch/extension.h>
+#include <torch/all.h>
 #include <ATen/cuda/CUDAContext.h>
 #include <c10/cuda/CUDAGuard.h>
+#include "../cuda_compat.h"
 
-#include <cub/cub.cuh>
+#ifndef USE_ROCM
     #include <cub/util_type.cuh>
+    #include <cub/cub.cuh>
+#else
+    #include <hipcub/util_type.hpp>
+    #include <hipcub/hipcub.hpp>
+#endif
+
+#define MAX(a, b) ((a) > (b) ? (a) : (b))
+#define MIN(a, b) ((a) < (b) ? (a) : (b))
 
 namespace vllm {
 namespace moe {
 
-static constexpr int WARP_SIZE = 32;
-
 /// Aligned array type
 template <
     typename T,
@@ -265,7 +272,7 @@ __launch_bounds__(WARPS_PER_CTA* WARP_SIZE) __global__
 #pragma unroll
     for (int mask = THREADS_PER_ROW / 2; mask > 0; mask /= 2)
     {
-        thread_max = max(thread_max, __shfl_xor_sync(0xFFFFFFFF, thread_max, mask, THREADS_PER_ROW));
+        thread_max = max(thread_max, VLLM_SHFL_XOR_SYNC_WIDTH(thread_max, mask, THREADS_PER_ROW));
     }
 
     // From this point, thread max in all the threads have the max within the row.
@@ -282,7 +289,7 @@ __launch_bounds__(WARPS_PER_CTA* WARP_SIZE) __global__
 #pragma unroll
     for (int mask = THREADS_PER_ROW / 2; mask > 0; mask /= 2)
     {
-        row_sum += __shfl_xor_sync(0xFFFFFFFF, row_sum, mask, THREADS_PER_ROW);
+        row_sum += VLLM_SHFL_XOR_SYNC_WIDTH(row_sum, mask, THREADS_PER_ROW);
     }
 
     // From this point, all threads have the max and the sum for their rows in the thread_max and thread_sum variables
@@ -332,8 +339,8 @@ __launch_bounds__(WARPS_PER_CTA* WARP_SIZE) __global__
 #pragma unroll
         for (int mask = THREADS_PER_ROW / 2; mask > 0; mask /= 2)
         {
-            float other_max = __shfl_xor_sync(0xFFFFFFFF, max_val, mask, THREADS_PER_ROW);
+            float other_max = VLLM_SHFL_XOR_SYNC_WIDTH(max_val, mask, THREADS_PER_ROW);
-            int other_expert = __shfl_xor_sync(0xFFFFFFFF, expert, mask, THREADS_PER_ROW);
+            int other_expert = VLLM_SHFL_XOR_SYNC_WIDTH(expert, mask, THREADS_PER_ROW);
 
             // We want lower indices to "win" in every thread so we break ties this way
             if (other_max > max_val || (other_max == max_val && other_expert < expert))
@@ -383,7 +390,7 @@ struct TopkConstants
 {
     static constexpr int ELTS_PER_LDG = BYTES_PER_LDG / sizeof(float);
     static_assert(EXPERTS / (ELTS_PER_LDG * WARP_SIZE) == 0 || EXPERTS % (ELTS_PER_LDG * WARP_SIZE) == 0, "");
-    static constexpr int VECs_PER_THREAD = std::max(1, EXPERTS / (ELTS_PER_LDG * WARP_SIZE));
+    static constexpr int VECs_PER_THREAD = MAX(1, EXPERTS / (ELTS_PER_LDG * WARP_SIZE));
     static constexpr int VPT = VECs_PER_THREAD * ELTS_PER_LDG;
     static constexpr int THREADS_PER_ROW = EXPERTS / VPT;
     static constexpr int ROWS_PER_WARP = WARP_SIZE / THREADS_PER_ROW;
@@ -396,7 +403,7 @@ void topkGatingSoftmaxLauncherHelper(const float* input, const bool* finished, f
 {
     static constexpr std::size_t MAX_BYTES_PER_LDG = 16;
 
-    static constexpr int BYTES_PER_LDG = std::min(MAX_BYTES_PER_LDG, sizeof(float) * EXPERTS);
+    static constexpr int BYTES_PER_LDG = MIN(MAX_BYTES_PER_LDG, sizeof(float) * EXPERTS);
     using Constants = detail::TopkConstants<EXPERTS, BYTES_PER_LDG>;
     static constexpr int VPT = Constants::VPT;
     static constexpr int ROWS_PER_WARP = Constants::ROWS_PER_WARP;

csrc/moe/torch_bindings.cpp

@@ -0,0 +1,12 @@
+#include "registration.h"
+#include "moe_ops.h"
+
+TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, m) {
+  // Apply topk softmax to the gating outputs.
+  m.def(
+      "topk_softmax(Tensor! topk_weights, Tensor! topk_indices, Tensor! "
+      "token_expert_indices, Tensor gating_output) -> ()");
+  m.impl("topk_softmax", torch::kCUDA, &topk_softmax);
+}
+
+REGISTER_EXTENSION(TORCH_EXTENSION_NAME)

csrc/moe_align_block_size_kernels.cu

@@ -1,4 +1,4 @@
-#include <torch/extension.h>
+#include <torch/all.h>
 #include <ATen/cuda/CUDAContext.h>
 
 #include <ATen/ATen.h>
@@ -12,11 +12,12 @@
 namespace vllm {
 
 namespace {
-__device__ __forceinline__ int32_t index(int32_t total_col, int32_t row, int32_t col) {
+__device__ __forceinline__ int32_t index(int32_t total_col, int32_t row,
+                                         int32_t col) {
   // don't worry about overflow because num_experts is relatively small
   return row * total_col + col;
 }
-}
+}  // namespace
 
 template <typename scalar_t>
 __global__ void moe_align_block_size_kernel(scalar_t* __restrict__ topk_ids,
@@ -24,15 +25,17 @@ __global__ void moe_align_block_size_kernel(scalar_t *__restrict__ topk_ids,
                                             int32_t* expert_ids,
                                             int32_t* total_tokens_post_pad,
                                             int32_t num_experts,
-                                int32_t block_size, 
+                                            int32_t block_size, size_t numel) {
-                                size_t numel) {
   const size_t tokens_per_thread = CEILDIV(numel, blockDim.x);
   const size_t start_idx = threadIdx.x * tokens_per_thread;
 
   extern __shared__ int32_t shared_mem[];
 
-    int32_t* tokens_cnts = shared_mem; // 2d tensor with shape (num_experts + 1, num_experts)
+  int32_t* tokens_cnts =
-    int32_t* cumsum = shared_mem + (num_experts + 1) * num_experts; // 1d tensor with shape (num_experts + 1)
+      shared_mem;  // 2d tensor with shape (num_experts + 1, num_experts)
+  int32_t* cumsum =
+      shared_mem + (num_experts + 1) *
+                       num_experts;  // 1d tensor with shape (num_experts + 1)
 
   for (int i = 0; i < num_experts; ++i) {
     tokens_cnts[index(num_experts, threadIdx.x + 1, i)] = 0;
@@ -40,8 +43,8 @@ __global__ void moe_align_block_size_kernel(scalar_t *__restrict__ topk_ids,
 
   /**
    * In the first step we compute token_cnts[thread_index + 1][expert_index],
-    * which counts how many tokens in the token shard of thread_index are assigned
+   * which counts how many tokens in the token shard of thread_index are
-    * to expert expert_index.
+   * assigned to expert expert_index.
    */
   for (int i = start_idx; i < numel && i < start_idx + tokens_per_thread; ++i) {
     ++tokens_cnts[index(num_experts, threadIdx.x + 1, topk_ids[i])];
@@ -52,7 +55,8 @@ __global__ void moe_align_block_size_kernel(scalar_t *__restrict__ topk_ids,
   // For each expert we accumulate the token counts from the different threads.
   tokens_cnts[index(num_experts, 0, threadIdx.x)] = 0;
   for (int i = 1; i <= blockDim.x; ++i) {
-        tokens_cnts[index(num_experts, i, threadIdx.x)] += tokens_cnts[index(num_experts, i-1, threadIdx.x)];
+    tokens_cnts[index(num_experts, i, threadIdx.x)] +=
+        tokens_cnts[index(num_experts, i - 1, threadIdx.x)];
   }
 
   __syncthreads();
@@ -61,7 +65,10 @@ __global__ void moe_align_block_size_kernel(scalar_t *__restrict__ topk_ids,
   if (threadIdx.x == 0) {
     cumsum[0] = 0;
     for (int i = 1; i <= num_experts; ++i) {
-            cumsum[i] = cumsum[i-1] + CEILDIV(tokens_cnts[index(num_experts, blockDim.x, i - 1)], block_size) * block_size;
+      cumsum[i] = cumsum[i - 1] +
+                  CEILDIV(tokens_cnts[index(num_experts, blockDim.x, i - 1)],
+                          block_size) *
+                      block_size;
     }
     *total_tokens_post_pad = cumsum[num_experts];
   }
@@ -69,57 +76,59 @@ __global__ void moe_align_block_size_kernel(scalar_t *__restrict__ topk_ids,
   __syncthreads();
 
   /**
-    * For each expert, each thread processes the tokens of the corresponding blocks
+   * For each expert, each thread processes the tokens of the corresponding
-    * and stores the corresponding expert_id for each block.
+   * blocks and stores the corresponding expert_id for each block.
    */
-    for (int i = cumsum[threadIdx.x];i < cumsum[threadIdx.x + 1];i += block_size) {
+  for (int i = cumsum[threadIdx.x]; i < cumsum[threadIdx.x + 1];
+       i += block_size) {
     expert_ids[i / block_size] = threadIdx.x;
   }
 
   /**
-    * Each thread processes a token shard, calculating the index of each token after
+   * Each thread processes a token shard, calculating the index of each token
-    * sorting by expert number. Given the example topk_ids = [0,1,2,1,2,3,0,3,4] and
+   * after sorting by expert number. Given the example topk_ids =
-    * block_size = 4, then the output would be [0, 6, *, *, 1, 3, *, *, 2, 4, *, *, 5, 7, *, *, 8, *, *, *],
+   * [0,1,2,1,2,3,0,3,4] and block_size = 4, then the output would be [0, 6, *,
-    * where * represents a padding value(preset in python).
+   * *, 1, 3, *, *, 2, 4, *, *, 5, 7, *, *, 8, *, *, *], where * represents a
+   * padding value(preset in python).
    */
   for (int i = start_idx; i < numel && i < start_idx + tokens_per_thread; ++i) {
     int32_t expert_id = topk_ids[i];
     /** The cumsum[expert_id] stores the starting index of the tokens that the
-        * expert with expert_id needs to process, and tokens_cnts[threadIdx.x][expert_id]
+     * expert with expert_id needs to process, and
-        * stores the indices of the tokens processed by the expert with expert_id within
+     * tokens_cnts[threadIdx.x][expert_id] stores the indices of the tokens
-        * the current thread's token shard.
+     * processed by the expert with expert_id within the current thread's token
+     * shard.
      */
-        int32_t rank_post_pad = tokens_cnts[index(num_experts, threadIdx.x, expert_id)] + cumsum[expert_id];
+    int32_t rank_post_pad =
+        tokens_cnts[index(num_experts, threadIdx.x, expert_id)] +
+        cumsum[expert_id];
     sorted_token_ids[rank_post_pad] = i;
     ++tokens_cnts[index(num_experts, threadIdx.x, expert_id)];
   }
 }
-}
+}  // namespace vllm
 
-void moe_align_block_size(
+void moe_align_block_size(torch::Tensor topk_ids, int64_t num_experts,
-    torch::Tensor topk_ids,
+                          int64_t block_size, torch::Tensor sorted_token_ids,
-    int num_experts,
-    int block_size,
-    torch::Tensor sorted_token_ids,
                           torch::Tensor experts_ids,
                           torch::Tensor num_tokens_post_pad) {
   const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
   VLLM_DISPATCH_INTEGRAL_TYPES(
       topk_ids.scalar_type(), "moe_align_block_size_kernel", [&] {
-        // calc needed amount of shared mem for `tokens_cnts` and `cumsum` tensors
+        // calc needed amount of shared mem for `tokens_cnts` and `cumsum`
-        const int32_t shared_mem = ((num_experts + 1) * num_experts + (num_experts + 1)) * sizeof(int32_t);
+        // tensors
+        const int32_t shared_mem =
+            ((num_experts + 1) * num_experts + (num_experts + 1)) *
+            sizeof(int32_t);
 
         // set dynamic shared mem
         auto kernel = vllm::moe_align_block_size_kernel<scalar_t>;
-        AT_CUDA_CHECK(
+        AT_CUDA_CHECK(VLLM_DevFuncAttribute_SET_MaxDynamicSharedMemorySize(
-            VLLM_DevFuncAttribute_SET_MaxDynamicSharedMemorySize((void *)kernel, shared_mem));
+            (void*)kernel, shared_mem));
         kernel<<<1, num_experts, shared_mem, stream>>>(
-            topk_ids.data_ptr<scalar_t>(),
+            topk_ids.data_ptr<scalar_t>(), sorted_token_ids.data_ptr<int32_t>(),
-            sorted_token_ids.data_ptr<int32_t>(), 
             experts_ids.data_ptr<int32_t>(),
-            num_tokens_post_pad.data_ptr<int32_t>(), 
+            num_tokens_post_pad.data_ptr<int32_t>(), num_experts, block_size,
-            num_experts,
-            block_size,
             topk_ids.numel());
       });
 }

csrc/ops.h

@@ -1,181 +1,146 @@
 #pragma once
 
-#include <torch/extension.h>
+#include <torch/library.h>
 
 void paged_attention_v1(
-  torch::Tensor& out,
+    torch::Tensor& out, torch::Tensor& query, torch::Tensor& key_cache,
-  torch::Tensor& query,
+    torch::Tensor& value_cache, int64_t num_kv_heads, double scale,
-  torch::Tensor& key_cache,
+    torch::Tensor& block_tables, torch::Tensor& seq_lens, int64_t block_size,
-  torch::Tensor& value_cache,
+    int64_t max_seq_len, const c10::optional<torch::Tensor>& alibi_slopes,
-  int num_kv_heads,
+    const std::string& kv_cache_dtype, double kv_scale, const int64_t tp_rank,
-  float scale,
+    const int64_t blocksparse_local_blocks,
-  torch::Tensor& block_tables,
+    const int64_t blocksparse_vert_stride, const int64_t blocksparse_block_size,
-  torch::Tensor& context_lens,
+    const int64_t blocksparse_head_sliding_step);
-  int block_size,
-  int max_context_len,
-  const c10::optional<torch::Tensor>& alibi_slopes,
-  const std::string& kv_cache_dtype,
-  float kv_scale);
 
 void paged_attention_v2(
-  torch::Tensor& out,
+    torch::Tensor& out, torch::Tensor& exp_sums, torch::Tensor& max_logits,
-  torch::Tensor& exp_sums,
+    torch::Tensor& tmp_out, torch::Tensor& query, torch::Tensor& key_cache,
-  torch::Tensor& max_logits,
+    torch::Tensor& value_cache, int64_t num_kv_heads, double scale,
-  torch::Tensor& tmp_out,
+    torch::Tensor& block_tables, torch::Tensor& seq_lens, int64_t block_size,
-  torch::Tensor& query,
+    int64_t max_seq_len, const c10::optional<torch::Tensor>& alibi_slopes,
-  torch::Tensor& key_cache,
+    const std::string& kv_cache_dtype, double kv_scale, const int64_t tp_rank,
-  torch::Tensor& value_cache,
+    const int64_t blocksparse_local_blocks,
-  int num_kv_heads,
+    const int64_t blocksparse_vert_stride, const int64_t blocksparse_block_size,
-  float scale,
+    const int64_t blocksparse_head_sliding_step);
-  torch::Tensor& block_tables,
-  torch::Tensor& context_lens,
-  int block_size,
-  int max_context_len,
-  const c10::optional<torch::Tensor>& alibi_slopes,
-  const std::string& kv_cache_dtype,
-  float kv_scale);
 
-void rms_norm(
+void rms_norm(torch::Tensor& out, torch::Tensor& input, torch::Tensor& weight,
-  torch::Tensor& out,
+              double epsilon);
-  torch::Tensor& input,
-  torch::Tensor& weight,
-  float epsilon);
 
-void fused_add_rms_norm(
+void fused_add_rms_norm(torch::Tensor& input, torch::Tensor& residual,
-  torch::Tensor& input,
+                        torch::Tensor& weight, double epsilon);
-  torch::Tensor& residual,
-  torch::Tensor& weight,
-  float epsilon);
 
-void rotary_embedding(
+void rotary_embedding(torch::Tensor& positions, torch::Tensor& query,
-  torch::Tensor& positions,
+                      torch::Tensor& key, int64_t head_size,
-  torch::Tensor& query,
+                      torch::Tensor& cos_sin_cache, bool is_neox);
-  torch::Tensor& key,
-  int head_size,
-  torch::Tensor& cos_sin_cache,
-  bool is_neox);
 
-void batched_rotary_embedding(
+void batched_rotary_embedding(torch::Tensor& positions, torch::Tensor& query,
-  torch::Tensor& positions,
+                              torch::Tensor& key, int64_t head_size,
-  torch::Tensor& query,
+                              torch::Tensor& cos_sin_cache, bool is_neox,
-  torch::Tensor& key,
+                              int64_t rot_dim,
-  int head_size,
-  torch::Tensor& cos_sin_cache,
-  bool is_neox,
-  int rot_dim,
                               torch::Tensor& cos_sin_cache_offsets);
 
-void silu_and_mul(
+void silu_and_mul(torch::Tensor& out, torch::Tensor& input);
-  torch::Tensor& out,
-  torch::Tensor& input);
 
-void gelu_and_mul(
+void gelu_and_mul(torch::Tensor& out, torch::Tensor& input);
-  torch::Tensor& out,
-  torch::Tensor& input);
 
-void gelu_tanh_and_mul(
+void gelu_tanh_and_mul(torch::Tensor& out, torch::Tensor& input);
-  torch::Tensor& out,
-  torch::Tensor& input);
 
-void gelu_new(
+void gelu_new(torch::Tensor& out, torch::Tensor& input);
-  torch::Tensor& out,
-  torch::Tensor& input);
 
-void gelu_fast(
+void gelu_fast(torch::Tensor& out, torch::Tensor& input);
-  torch::Tensor& out,
-  torch::Tensor& input);
 
 #ifndef USE_ROCM
-torch::Tensor aqlm_gemm(
+torch::Tensor aqlm_gemm(const torch::Tensor& input, const torch::Tensor& codes,
-  const torch::Tensor& input,
-  const torch::Tensor& codes,
                         const torch::Tensor& codebooks,
                         const torch::Tensor& scales,
                         const torch::Tensor& codebook_partition_sizes,
-  const std::optional<torch::Tensor>& bias
+                        const std::optional<torch::Tensor>& bias);
-);
 
-torch::Tensor aqlm_dequant(
+torch::Tensor aqlm_dequant(const torch::Tensor& codes,
-  const torch::Tensor& codes,
                            const torch::Tensor& codebooks,
-  const torch::Tensor& codebook_partition_sizes
+                           const torch::Tensor& codebook_partition_sizes);
-);
 
-torch::Tensor awq_gemm(
+torch::Tensor awq_gemm(torch::Tensor _in_feats, torch::Tensor _kernel,
-  torch::Tensor _in_feats,
+                       torch::Tensor _scaling_factors, torch::Tensor _zeros,
-  torch::Tensor _kernel,
+                       int64_t split_k_iters);
+
+torch::Tensor awq_dequantize(torch::Tensor _kernel,
                              torch::Tensor _scaling_factors,
-  torch::Tensor _zeros,
+                             torch::Tensor _zeros, int64_t split_k_iters,
-  int split_k_iters);
+                             int64_t thx, int64_t thy);
 
-torch::Tensor awq_dequantize(
+torch::Tensor marlin_gemm(torch::Tensor& a, torch::Tensor& b_q_weight,
-    torch::Tensor _kernel,
+                          torch::Tensor& b_scales, torch::Tensor& workspace,
-    torch::Tensor _scaling_factors,
+                          int64_t size_m, int64_t size_n, int64_t size_k);
-    torch::Tensor _zeros,
-    int split_k_iters,
-    int thx,
-    int thy);
 
-torch::Tensor marlin_gemm(
+torch::Tensor gptq_marlin_24_gemm(torch::Tensor& a, torch::Tensor& b_q_weight,
-    torch::Tensor& a, 
+                                  torch::Tensor& b_meta,
-    torch::Tensor& b_q_weight,
                                   torch::Tensor& b_scales,
-    torch::Tensor& workspace,
+                                  torch::Tensor& workspace, int64_t num_bits,
-    int64_t size_m, 
+                                  int64_t size_m, int64_t size_n,
-    int64_t size_n, 
                                   int64_t size_k);
+
+torch::Tensor gptq_marlin_gemm(torch::Tensor& a, torch::Tensor& b_q_weight,
+                               torch::Tensor& b_scales, torch::Tensor& g_idx,
+                               torch::Tensor& perm, torch::Tensor& workspace,
+                               int64_t num_bits, int64_t size_m, int64_t size_n,
+                               int64_t size_k, bool is_k_full);
+
+torch::Tensor gptq_marlin_repack(torch::Tensor& b_q_weight, torch::Tensor& perm,
+                                 int64_t size_k, int64_t size_n,
+                                 int64_t num_bits);
+
+void cutlass_scaled_mm_dq(torch::Tensor& out, torch::Tensor const& a,
+                          torch::Tensor const& b, torch::Tensor const& a_scales,
+                          torch::Tensor const& b_scales);
+
 #endif
 
-void squeezellm_gemm(
+void static_scaled_int8_quant(torch::Tensor& out, torch::Tensor const& input,
-  torch::Tensor vec,
+                              torch::Tensor const& scale);
-  torch::Tensor mat,
+
-  torch::Tensor mul,
+void dynamic_scaled_int8_quant(torch::Tensor& out, torch::Tensor const& input,
+                               torch::Tensor& scales);
+
+void squeezellm_gemm(torch::Tensor vec, torch::Tensor mat, torch::Tensor mul,
                      torch::Tensor lookup_table);
 
-torch::Tensor gptq_gemm(
+torch::Tensor gptq_gemm(torch::Tensor a, torch::Tensor b_q_weight,
-  torch::Tensor a,
-  torch::Tensor b_q_weight,
                         torch::Tensor b_gptq_qzeros,
-  torch::Tensor b_gptq_scales,
+                        torch::Tensor b_gptq_scales, torch::Tensor b_g_idx,
-  torch::Tensor b_g_idx,
+                        bool use_exllama, int64_t bit);
-  bool use_exllama,
-  int bit);
 
-void gptq_shuffle(
+void gptq_shuffle(torch::Tensor q_weight, torch::Tensor q_perm, int64_t bit);
-  torch::Tensor q_weight,
-  torch::Tensor q_perm,
-  int bit);
 
-void scaled_fp8_quant(
+void static_scaled_fp8_quant(torch::Tensor& out, torch::Tensor& input,
-  torch::Tensor& out,
-  torch::Tensor& input,
                              torch::Tensor& scale);
 
-void moe_align_block_size(
+void dynamic_scaled_fp8_quant(torch::Tensor& out, torch::Tensor& input,
-  torch::Tensor topk_ids,
+                              torch::Tensor& scale);
-  int num_experts,
+
-  int block_size,
+void moe_align_block_size(torch::Tensor topk_ids, int64_t num_experts,
-  torch::Tensor sorted_token_ids,
+                          int64_t block_size, torch::Tensor sorted_token_ids,
                           torch::Tensor experts_ids,
                           torch::Tensor num_tokens_post_pad);
 
 #ifndef USE_ROCM
-using fptr_t = uint64_t;
+using fptr_t = int64_t;
 fptr_t init_custom_ar(torch::Tensor& meta, torch::Tensor& rank_data,
                       const std::vector<std::string>& handles,
-                    const std::vector<int64_t> &offsets, int rank,
+                      const std::vector<int64_t>& offsets, int64_t rank,
                       bool full_nvlink);
-bool should_custom_ar(torch::Tensor &inp, int max_size, int world_size,
+bool should_custom_ar(torch::Tensor& inp, int64_t max_size, int64_t world_size,
                       bool full_nvlink);
 void all_reduce_reg(fptr_t _fa, torch::Tensor& inp, torch::Tensor& out);
 void all_reduce_unreg(fptr_t _fa, torch::Tensor& inp, torch::Tensor& reg_buffer,
                       torch::Tensor& out);
 void dispose(fptr_t _fa);
-int meta_size();
+int64_t meta_size();
 void register_buffer(fptr_t _fa, torch::Tensor& t,
                      const std::vector<std::string>& handles,
                      const std::vector<int64_t>& offsets);
-std::pair<std::vector<uint8_t>, std::vector<int64_t>> get_graph_buffer_ipc_meta(fptr_t _fa);
+std::tuple<torch::Tensor, std::vector<int64_t>> get_graph_buffer_ipc_meta(
+    fptr_t _fa);
 void register_graph_buffers(fptr_t _fa, const std::vector<std::string>& handles,
                             const std::vector<std::vector<int64_t>>& offsets);
 #endif

csrc/pos_encoding_kernels.cu

@@ -1,4 +1,4 @@
-#include <torch/extension.h>
+#include <torch/all.h>
 #include <ATen/cuda/CUDAContext.h>
 #include <c10/cuda/CUDAGuard.h>
 
@@ -9,12 +9,8 @@ namespace vllm {
 
 template <typename scalar_t, bool IS_NEOX>
 inline __device__ void apply_token_rotary_embedding(
-  scalar_t* __restrict__ arr,
+    scalar_t* __restrict__ arr, const scalar_t* __restrict__ cos_ptr,
-  const scalar_t* __restrict__ cos_ptr,
+    const scalar_t* __restrict__ sin_ptr, int rot_offset, int embed_dim) {
-  const scalar_t* __restrict__ sin_ptr,
-  int rot_offset,
-  int embed_dim)
-{
   int x_index, y_index;
   scalar_t cos, sin;
   if (IS_NEOX) {
@@ -39,17 +35,15 @@ inline __device__ void apply_token_rotary_embedding(
 
 template <typename scalar_t, bool IS_NEOX>
 inline __device__ void apply_rotary_embedding(
-  scalar_t* __restrict__ query,                 // [batch_size, seq_len, num_heads, head_size] or [num_tokens, num_heads, head_size]
+    scalar_t* __restrict__ query,  // [batch_size, seq_len, num_heads,
-  scalar_t* __restrict__ key,                   // [batch_size, seq_len, num_kv_heads, head_size] or [num_tokens, num_kv_heads, head_size]
+                                   // head_size] or [num_tokens, num_heads,
-  const scalar_t* cache_ptr,
+                                   // head_size]
-  const int head_size,
+    scalar_t* __restrict__ key,    // [batch_size, seq_len, num_kv_heads,
-  const int num_heads,
+                                   // head_size] or [num_tokens, num_kv_heads,
-  const int num_kv_heads,
+                                   // head_size]
-  const int rot_dim,
+    const scalar_t* cache_ptr, const int head_size, const int num_heads,
-  const int token_idx,
+    const int num_kv_heads, const int rot_dim, const int token_idx,
-  const int64_t query_stride,
+    const int64_t query_stride, const int64_t key_stride) {
-  const int64_t key_stride)
-{
   const int embed_dim = rot_dim / 2;
   const scalar_t* cos_ptr = cache_ptr;
   const scalar_t* sin_ptr = cache_ptr + embed_dim;
@@ -59,8 +53,8 @@ inline __device__ void apply_rotary_embedding(
     const int head_idx = i / embed_dim;
     const int64_t token_head = token_idx * query_stride + head_idx * head_size;
     const int rot_offset = i % embed_dim;
-    apply_token_rotary_embedding<scalar_t, IS_NEOX>(query + token_head, cos_ptr,
+    apply_token_rotary_embedding<scalar_t, IS_NEOX>(
-                                              sin_ptr, rot_offset, embed_dim);
+        query + token_head, cos_ptr, sin_ptr, rot_offset, embed_dim);
   }
 
   const int nk = num_kv_heads * embed_dim;
@@ -68,60 +62,72 @@ inline __device__ void apply_rotary_embedding(
     const int head_idx = i / embed_dim;
     const int64_t token_head = token_idx * key_stride + head_idx * head_size;
     const int rot_offset = i % embed_dim;
-    apply_token_rotary_embedding<scalar_t, IS_NEOX>(key + token_head, cos_ptr,
+    apply_token_rotary_embedding<scalar_t, IS_NEOX>(
-                                              sin_ptr, rot_offset, embed_dim);
+        key + token_head, cos_ptr, sin_ptr, rot_offset, embed_dim);
   }
 }
 
 template <typename scalar_t, bool IS_NEOX>
 __global__ void rotary_embedding_kernel(
-  const int64_t* __restrict__ positions,        // [batch_size, seq_len] or [num_tokens]
+    const int64_t* __restrict__ positions,  // [batch_size, seq_len] or
-  scalar_t* __restrict__ query,                 // [batch_size, seq_len, num_heads, head_size] or [num_tokens, num_heads, head_size]
+                                            // [num_tokens]
-  scalar_t* __restrict__ key,                   // [batch_size, seq_len, num_kv_heads, head_size] or [num_tokens, num_kv_heads, head_size]
+    scalar_t* __restrict__ query,           // [batch_size, seq_len, num_heads,
-  const scalar_t* __restrict__ cos_sin_cache,   // [max_position, 2, rot_dim // 2]
+                                   // head_size] or [num_tokens, num_heads,
-  const int rot_dim,
+                                   // head_size]
-  const int64_t query_stride,
+    scalar_t* __restrict__ key,  // [batch_size, seq_len, num_kv_heads,
-  const int64_t key_stride,
+                                 // head_size] or [num_tokens, num_kv_heads,
-  const int num_heads,
+                                 // head_size]
-  const int num_kv_heads,
+    const scalar_t* __restrict__ cos_sin_cache,  // [max_position, 2, rot_dim //
-  const int head_size) {
+                                                 // 2]
+    const int rot_dim, const int64_t query_stride, const int64_t key_stride,
+    const int num_heads, const int num_kv_heads, const int head_size) {
   // Each thread block is responsible for one token.
   const int token_idx = blockIdx.x;
   int64_t pos = positions[token_idx];
   const scalar_t* cache_ptr = cos_sin_cache + pos * rot_dim;
 
-  apply_rotary_embedding<scalar_t, IS_NEOX>(query, key, cache_ptr, head_size, num_heads, num_kv_heads, rot_dim, token_idx, query_stride, key_stride);
+  apply_rotary_embedding<scalar_t, IS_NEOX>(
+      query, key, cache_ptr, head_size, num_heads, num_kv_heads, rot_dim,
+      token_idx, query_stride, key_stride);
 }
 
 template <typename scalar_t, bool IS_NEOX>
 __global__ void batched_rotary_embedding_kernel(
-  const int64_t* __restrict__ positions,              // [batch_size, seq_len] or [num_tokens]
+    const int64_t* __restrict__ positions,  // [batch_size, seq_len] or
-  scalar_t* __restrict__ query,                       // [batch_size, seq_len, num_heads, head_size] or [num_tokens, num_heads, head_size]
+                                            // [num_tokens]
-  scalar_t* __restrict__ key,                         // [batch_size, seq_len, num_kv_heads, head_size] or [num_tokens, num_kv_heads, head_size]
+    scalar_t* __restrict__ query,           // [batch_size, seq_len, num_heads,
-  const scalar_t* __restrict__ cos_sin_cache,         // [max_position, 2, rot_dim // 2]
+                                   // head_size] or [num_tokens, num_heads,
-  const int64_t* __restrict__ cos_sin_cache_offsets,  // [batch_size, seq_len] or [num_tokens]
+                                   // head_size]
-  const int rot_dim,
+    scalar_t* __restrict__ key,  // [batch_size, seq_len, num_kv_heads,
-  const int64_t query_stride,
+                                 // head_size] or [num_tokens, num_kv_heads,
-  const int64_t key_stride,
+                                 // head_size]
-  const int num_heads,
+    const scalar_t* __restrict__ cos_sin_cache,  // [max_position, 2, rot_dim //
-  const int num_kv_heads,
+                                                 // 2]
-  const int head_size) {
+    const int64_t* __restrict__ cos_sin_cache_offsets,  // [batch_size, seq_len]
+                                                        // or [num_tokens]
+    const int rot_dim, const int64_t query_stride, const int64_t key_stride,
+    const int num_heads, const int num_kv_heads, const int head_size) {
   // Each thread block is responsible for one token.
   const int token_idx = blockIdx.x;
   int64_t pos = positions[token_idx];
   int64_t cos_sin_cache_offset = cos_sin_cache_offsets[token_idx];
-  const scalar_t* cache_ptr = cos_sin_cache + (cos_sin_cache_offset + pos) * rot_dim;
+  const scalar_t* cache_ptr =
+      cos_sin_cache + (cos_sin_cache_offset + pos) * rot_dim;
 
-  apply_rotary_embedding<scalar_t, IS_NEOX>(query, key, cache_ptr, head_size, num_heads, num_kv_heads, rot_dim, token_idx, query_stride, key_stride);
+  apply_rotary_embedding<scalar_t, IS_NEOX>(
+      query, key, cache_ptr, head_size, num_heads, num_kv_heads, rot_dim,
+      token_idx, query_stride, key_stride);
 }
 
 }  // namespace vllm
 
 void rotary_embedding(
     torch::Tensor& positions,  // [batch_size, seq_len] or [num_tokens]
-  torch::Tensor& query,             // [batch_size, seq_len, num_heads * head_size] or [num_tokens, num_heads * head_size]
+    torch::Tensor& query,  // [batch_size, seq_len, num_heads * head_size] or
-  torch::Tensor& key,               // [batch_size, seq_len, num_kv_heads * head_size] or [num_tokens, num_kv_heads * head_size]
+                           // [num_tokens, num_heads * head_size]
-  int head_size,
+    torch::Tensor& key,    // [batch_size, seq_len, num_kv_heads * head_size] or
+                           // [num_tokens, num_kv_heads * head_size]
+    int64_t head_size,
     torch::Tensor& cos_sin_cache,  // [max_position, rot_dim]
     bool is_neox) {
   int64_t num_tokens = query.numel() / query.size(-1);
@@ -132,36 +138,21 @@ void rotary_embedding(
   int64_t key_stride = key.stride(-2);
 
   dim3 grid(num_tokens);
-  dim3 block(std::min(num_heads * rot_dim / 2, 512));
+  dim3 block(std::min<int64_t>(num_heads * rot_dim / 2, 512));
   const at::cuda::OptionalCUDAGuard device_guard(device_of(query));
   const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
-  VLLM_DISPATCH_FLOATING_TYPES(
+  VLLM_DISPATCH_FLOATING_TYPES(query.scalar_type(), "rotary_embedding", [&] {
-    query.scalar_type(),
-    "rotary_embedding",
-    [&] {
     if (is_neox) {
       vllm::rotary_embedding_kernel<scalar_t, true><<<grid, block, 0, stream>>>(
-          positions.data_ptr<int64_t>(),
+          positions.data_ptr<int64_t>(), query.data_ptr<scalar_t>(),
-          query.data_ptr<scalar_t>(),
+          key.data_ptr<scalar_t>(), cos_sin_cache.data_ptr<scalar_t>(), rot_dim,
-          key.data_ptr<scalar_t>(),
+          query_stride, key_stride, num_heads, num_kv_heads, head_size);
-          cos_sin_cache.data_ptr<scalar_t>(),
-          rot_dim,
-          query_stride,
-          key_stride,
-          num_heads,
-          num_kv_heads,
-          head_size);
     } else {
-        vllm::rotary_embedding_kernel<scalar_t, false><<<grid, block, 0, stream>>>(
+      vllm::rotary_embedding_kernel<scalar_t, false>
-          positions.data_ptr<int64_t>(),
+          <<<grid, block, 0, stream>>>(
-          query.data_ptr<scalar_t>(),
+              positions.data_ptr<int64_t>(), query.data_ptr<scalar_t>(),
-          key.data_ptr<scalar_t>(),
+              key.data_ptr<scalar_t>(), cos_sin_cache.data_ptr<scalar_t>(),
-          cos_sin_cache.data_ptr<scalar_t>(),
+              rot_dim, query_stride, key_stride, num_heads, num_kv_heads,
-          rot_dim,
-          query_stride,
-          key_stride,
-          num_heads,
-          num_kv_heads,
               head_size);
     }
   });
@@ -173,12 +164,13 @@ and process in batched manner.
 */
 void batched_rotary_embedding(
     torch::Tensor& positions,  // [batch_size, seq_len] or [num_tokens]
-  torch::Tensor& query,             // [batch_size, seq_len, num_heads * head_size] or [num_tokens, num_heads * head_size]
+    torch::Tensor& query,  // [batch_size, seq_len, num_heads * head_size] or
-  torch::Tensor& key,               // [batch_size, seq_len, num_kv_heads * head_size] or [num_tokens, num_kv_heads * head_size]
+                           // [num_tokens, num_heads * head_size]
-  int head_size,
+    torch::Tensor& key,    // [batch_size, seq_len, num_kv_heads * head_size] or
+                           // [num_tokens, num_kv_heads * head_size]
+    int64_t head_size,
     torch::Tensor& cos_sin_cache,  // [max_position, rot_dim]
-  bool is_neox,
+    bool is_neox, int64_t rot_dim,
-  int rot_dim,
     torch::Tensor& cos_sin_cache_offsets  // [num_tokens]
 ) {
   int64_t num_tokens = cos_sin_cache_offsets.size(0);
@@ -188,39 +180,24 @@ void batched_rotary_embedding(
   int64_t key_stride = key.stride(-2);
 
   dim3 grid(num_tokens);
-  dim3 block(std::min(num_heads * rot_dim / 2, 512));
+  dim3 block(std::min<int64_t>(num_heads * rot_dim / 2, 512));
   const at::cuda::OptionalCUDAGuard device_guard(device_of(query));
   const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
-  VLLM_DISPATCH_FLOATING_TYPES(
+  VLLM_DISPATCH_FLOATING_TYPES(query.scalar_type(), "rotary_embedding", [&] {
-    query.scalar_type(),
-    "rotary_embedding",
-    [&] {
     if (is_neox) {
-        vllm::batched_rotary_embedding_kernel<scalar_t, true><<<grid, block, 0, stream>>>(
+      vllm::batched_rotary_embedding_kernel<scalar_t, true>
-          positions.data_ptr<int64_t>(),
+          <<<grid, block, 0, stream>>>(
-          query.data_ptr<scalar_t>(),
+              positions.data_ptr<int64_t>(), query.data_ptr<scalar_t>(),
-          key.data_ptr<scalar_t>(),
+              key.data_ptr<scalar_t>(), cos_sin_cache.data_ptr<scalar_t>(),
-          cos_sin_cache.data_ptr<scalar_t>(),
+              cos_sin_cache_offsets.data_ptr<int64_t>(), rot_dim, query_stride,
-          cos_sin_cache_offsets.data_ptr<int64_t>(),
+              key_stride, num_heads, num_kv_heads, head_size);
-          rot_dim,
-          query_stride,
-          key_stride,
-          num_heads,
-          num_kv_heads,
-          head_size);
     } else {
-        vllm::batched_rotary_embedding_kernel<scalar_t, false><<<grid, block, 0, stream>>>(
+      vllm::batched_rotary_embedding_kernel<scalar_t, false>
-          positions.data_ptr<int64_t>(),
+          <<<grid, block, 0, stream>>>(
-          query.data_ptr<scalar_t>(),
+              positions.data_ptr<int64_t>(), query.data_ptr<scalar_t>(),
-          key.data_ptr<scalar_t>(),
+              key.data_ptr<scalar_t>(), cos_sin_cache.data_ptr<scalar_t>(),
-          cos_sin_cache.data_ptr<scalar_t>(),
+              cos_sin_cache_offsets.data_ptr<int64_t>(), rot_dim, query_stride,
-          cos_sin_cache_offsets.data_ptr<int64_t>(),
+              key_stride, num_heads, num_kv_heads, head_size);
-          rot_dim,
-          query_stride,
-          key_stride,
-          num_heads,
-          num_kv_heads,
-          head_size);
     }
   });
 }

csrc/punica/bgmv/bgmv_bf16_bf16_bf16.cu

@@ -2,3 +2,4 @@
 #include "bgmv_impl.cuh"
 
 FOR_BGMV_WIDE_NARROW(INST_BGMV_TWOSIDE, nv_bfloat16, nv_bfloat16, nv_bfloat16)
+FOR_INST_BGMV_WIDE_NARROW(INST_BGMV_ONESIDE, nv_bfloat16, nv_bfloat16, nv_bfloat16)

csrc/punica/bgmv/bgmv_bf16_fp32_bf16.cu

@@ -2,3 +2,4 @@
 #include "bgmv_impl.cuh"
 
 FOR_BGMV_WIDE_NARROW(INST_BGMV_TWOSIDE, nv_bfloat16, float, nv_bfloat16)
+FOR_INST_BGMV_WIDE_NARROW(INST_BGMV_ONESIDE, nv_bfloat16, float, nv_bfloat16)

csrc/punica/bgmv/bgmv_config.h

@@ -28,6 +28,7 @@ void bgmv_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
     f(in_T, out_T, W_T, narrow, 2752) \
     f(in_T, out_T, W_T, narrow, 2816) \
     f(in_T, out_T, W_T, narrow, 3072) \
+    f(in_T, out_T, W_T, narrow, 3328) \
     f(in_T, out_T, W_T, narrow, 3456) \
     f(in_T, out_T, W_T, narrow, 3584) \
     f(in_T, out_T, W_T, narrow, 4096) \
@@ -36,6 +37,7 @@ void bgmv_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
     f(in_T, out_T, W_T, narrow, 5504) \
     f(in_T, out_T, W_T, narrow, 5632) \
     f(in_T, out_T, W_T, narrow, 6144) \
+    f(in_T, out_T, W_T, narrow, 6400) \
     f(in_T, out_T, W_T, narrow, 6848) \
     f(in_T, out_T, W_T, narrow, 6912) \
     f(in_T, out_T, W_T, narrow, 7168) \
@@ -53,6 +55,7 @@ void bgmv_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
     f(in_T, out_T, W_T, narrow, 22016) \
     f(in_T, out_T, W_T, narrow, 24576) \
     f(in_T, out_T, W_T, narrow, 27392) \
+    f(in_T, out_T, W_T, narrow, 27648) \
     f(in_T, out_T, W_T, narrow, 28672) \
     f(in_T, out_T, W_T, narrow, 32000) \
     f(in_T, out_T, W_T, narrow, 32256) \
@@ -74,6 +77,77 @@ void bgmv_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
 // Keep above in sync with vllm/lora/layers::LogitsProcessorWithLoRA
 // and vllm/tests/lora/test_punica.py
 
+// Used for defining kernels going from the variety of 
+// dim in to the narrow dim out
+    // Using it for the fully sharded column 
+    // parallel LoRA A which splits the rank dim
+#define FOR_INST_BGMV_NARROW(f, in_T, out_T, W_T, narrow) \
+    f(in_T, out_T, W_T, 128, narrow) \
+    f(in_T, out_T, W_T, 256, narrow) \
+    f(in_T, out_T, W_T, 512, narrow) \
+    f(in_T, out_T, W_T, 640, narrow) \
+    f(in_T, out_T, W_T, 768, narrow) \
+    f(in_T, out_T, W_T, 1024, narrow) \
+    f(in_T, out_T, W_T, 1152, narrow) \
+    f(in_T, out_T, W_T, 1280, narrow) \
+    f(in_T, out_T, W_T, 1536, narrow) \
+    f(in_T, out_T, W_T, 1728, narrow) \
+    f(in_T, out_T, W_T, 1792, narrow) \
+    f(in_T, out_T, W_T, 2048, narrow) \
+    f(in_T, out_T, W_T, 2304, narrow) \
+    f(in_T, out_T, W_T, 2560, narrow) \
+    f(in_T, out_T, W_T, 2752, narrow) \
+    f(in_T, out_T, W_T, 2816, narrow) \
+    f(in_T, out_T, W_T, 3072, narrow) \
+    f(in_T, out_T, W_T, 3328, narrow) \
+    f(in_T, out_T, W_T, 3456, narrow) \
+    f(in_T, out_T, W_T, 3584, narrow) \
+    f(in_T, out_T, W_T, 4096, narrow) \
+    f(in_T, out_T, W_T, 4608, narrow) \
+    f(in_T, out_T, W_T, 5120, narrow) \
+    f(in_T, out_T, W_T, 5504, narrow) \
+    f(in_T, out_T, W_T, 5632, narrow) \
+    f(in_T, out_T, W_T, 6144, narrow) \
+    f(in_T, out_T, W_T, 6400, narrow) \
+    f(in_T, out_T, W_T, 6848, narrow) \
+    f(in_T, out_T, W_T, 6912, narrow) \
+    f(in_T, out_T, W_T, 7168, narrow) \
+    f(in_T, out_T, W_T, 8192, narrow) \
+    f(in_T, out_T, W_T, 9216, narrow) \
+    f(in_T, out_T, W_T, 10240, narrow) \
+    f(in_T, out_T, W_T, 11008, narrow) \
+    f(in_T, out_T, W_T, 12288, narrow) \
+    f(in_T, out_T, W_T, 13696, narrow) \
+    f(in_T, out_T, W_T, 13824, narrow) \
+    f(in_T, out_T, W_T, 14336, narrow) \
+    f(in_T, out_T, W_T, 15360, narrow) \
+    f(in_T, out_T, W_T, 16384, narrow) \
+    f(in_T, out_T, W_T, 20480, narrow) \
+    f(in_T, out_T, W_T, 22016, narrow) \
+    f(in_T, out_T, W_T, 24576, narrow) \
+    f(in_T, out_T, W_T, 27392, narrow) \
+    f(in_T, out_T, W_T, 27648, narrow) \
+    f(in_T, out_T, W_T, 28672, narrow) \
+    f(in_T, out_T, W_T, 32000, narrow) \
+    f(in_T, out_T, W_T, 32256, narrow) \
+    f(in_T, out_T, W_T, 32512, narrow) \
+    f(in_T, out_T, W_T, 32768, narrow) \
+    f(in_T, out_T, W_T, 33024, narrow) \
+    f(in_T, out_T, W_T, 36864, narrow) \
+    f(in_T, out_T, W_T, 43264, narrow) \
+    f(in_T, out_T, W_T, 49152, narrow) \
+    f(in_T, out_T, W_T, 64000, narrow) \
+    f(in_T, out_T, W_T, 64256, narrow) \
+    f(in_T, out_T, W_T, 64512, narrow) \
+    f(in_T, out_T, W_T, 102400, narrow) \
+    f(in_T, out_T, W_T, 102656, narrow) \
+    f(in_T, out_T, W_T, 102912, narrow) \
+    f(in_T, out_T, W_T, 128000, narrow) \
+    f(in_T, out_T, W_T, 128256, narrow) \
+    f(in_T, out_T, W_T, 128512, narrow) \
+// Keep above in sync with vllm/lora/layers::SamplerWithLoRA
+
+
 // Keep this in sync with vllm/config::LoRAConfig
 #define FOR_BGMV_WIDE_NARROW(f, in_T, out_T, W_T) \
     FOR_BGMV_WIDE(f, in_T, out_T, W_T, 8)  \
@@ -81,4 +155,14 @@ void bgmv_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
     FOR_BGMV_WIDE(f, in_T, out_T, W_T, 32) \
     FOR_BGMV_WIDE(f, in_T, out_T, W_T, 64)
 
+
+#define FOR_INST_BGMV_WIDE_NARROW(f, in_T, out_T, W_T) \
+    FOR_INST_BGMV_NARROW(f, in_T, out_T, W_T, 1) \
+    FOR_INST_BGMV_NARROW(f, in_T, out_T, W_T, 2) \
+    FOR_INST_BGMV_NARROW(f, in_T, out_T, W_T, 4) \
+    f(in_T, out_T, W_T, 8, 64) \
+    f(in_T, out_T, W_T, 16, 64) \
+    f(in_T, out_T, W_T, 32, 64) \
+    f(in_T, out_T, W_T, 64, 64)
+
 // clang-format on

csrc/punica/bgmv/bgmv_fp16_fp16_fp16.cu

@@ -2,3 +2,4 @@
 #include "bgmv_impl.cuh"
 
 FOR_BGMV_WIDE_NARROW(INST_BGMV_TWOSIDE, nv_half, nv_half, nv_half)
+FOR_INST_BGMV_WIDE_NARROW(INST_BGMV_ONESIDE, nv_half, nv_half, nv_half)

csrc/punica/bgmv/bgmv_fp16_fp32_fp16.cu

@@ -2,3 +2,4 @@
 #include "bgmv_impl.cuh"
 
 FOR_BGMV_WIDE_NARROW(INST_BGMV_TWOSIDE, nv_half, float, nv_half)
+FOR_INST_BGMV_WIDE_NARROW(INST_BGMV_ONESIDE, nv_half, float, nv_half)

csrc/punica/bgmv/bgmv_fp32_bf16_bf16.cu

@@ -2,3 +2,4 @@
 #include "bgmv_impl.cuh"
 
 FOR_BGMV_WIDE_NARROW(INST_BGMV_TWOSIDE, float, nv_bfloat16, nv_bfloat16)
+FOR_INST_BGMV_WIDE_NARROW(INST_BGMV_ONESIDE, float, nv_bfloat16, nv_bfloat16)

csrc/punica/bgmv/bgmv_fp32_fp16_fp16.cu

@@ -2,3 +2,4 @@
 #include "bgmv_impl.cuh"
 
 FOR_BGMV_WIDE_NARROW(INST_BGMV_TWOSIDE, float, nv_half, nv_half)
+FOR_INST_BGMV_WIDE_NARROW(INST_BGMV_ONESIDE, float, nv_half, nv_half)

csrc/punica/bgmv/bgmv_impl.cuh

@@ -1,8 +1,14 @@
 #pragma once
 
 #include <ATen/cuda/CUDAContext.h>
+#ifndef USE_ROCM
 #include <cooperative_groups.h>
+#else
+#include <hip/hip_cooperative_groups.h>
+#endif
+#ifndef USE_ROCM
 #include <cuda/pipeline>
+#endif
 #include <cuda_runtime.h>
 #include <iostream>
 #include <stdio.h>
@@ -11,6 +17,24 @@
 
 namespace cg = cooperative_groups;
 
+#ifdef USE_ROCM
+template <size_t len>
+__host__ __device__
+inline void* memcpy_blocking(void *dst, const void *src) {
+  // Does not handle the case of long datatypes
+  char *d = reinterpret_cast<char *>(dst);
+  const char *s = reinterpret_cast<const char *>(src);
+  size_t i = 0;
+#pragma unroll
+  for (i = 0; i < len; ++i) {
+    d[i] = s[i];
+  }
+  return dst;
+}
+#endif
+
+#ifndef USE_ROCM
+
 // nthrs = (32, 4)
 template <int feat_in, int feat_out, size_t vec_size, size_t X_copy_size,
           size_t W_copy_size, int tx, int ty, int tz, typename in_T,
@@ -141,6 +165,81 @@ bgmv_shrink_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
   }
 }
 
+#else
+
+template <int feat_in, int feat_out, size_t vec_size, size_t X_copy_size,
+          size_t W_copy_size, int tx, int ty, int tz, typename in_T,
+          typename out_T, typename W_T>
+__global__ void
+bgmv_shrink_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
+                   const W_T *__restrict__ W,
+                   const int64_t *__restrict__ indicies, int64_t y_offset,
+                   int64_t full_y_size, int64_t num_layers, int64_t layer_idx,
+                   float scale) {
+  size_t batch_idx = blockIdx.y;
+  int64_t idx = indicies[batch_idx] * num_layers + layer_idx;
+  if (idx < 0) {
+    return;
+  }
+
+  size_t j = blockIdx.x;
+  constexpr size_t tile_size = tx * ty * vec_size;
+  constexpr size_t num_tiles = (feat_in + tile_size - 1) / tile_size;
+  __shared__ float y_warpwise[ty];
+
+  float y = 0;
+  vec_t<in_T, vec_size> x_vec;
+  vec_t<W_T, vec_size> w_vec;
+  size_t tile_idx;
+
+#pragma unroll
+  for (tile_idx = 0; tile_idx < num_tiles; ++tile_idx) {
+    if (tile_idx * tile_size + (threadIdx.y * tx + threadIdx.x + 1) * vec_size - 1 < feat_in) {
+      x_vec.load(X + (batch_idx * feat_in) +
+                     tile_idx * tile_size +
+                     (threadIdx.y * tx + threadIdx.x) * vec_size);
+      w_vec.load(W + (idx * feat_out + j) * feat_in +
+                     tile_idx * tile_size +
+                     (threadIdx.y * tx + threadIdx.x) * vec_size);
+    }
+
+    float sum = 0.f;
+#pragma unroll
+    for (size_t i = 0; i < vec_size; ++i) {
+      sum += convert_type<W_T, float>(w_vec[i]) * convert_type<in_T, float>(x_vec[i]) * scale;
+    }
+#pragma unroll
+    for (size_t offset = tx / 2; offset > 0; offset /= 2) {
+      sum += VLLM_SHFL_DOWN_SYNC(sum, offset);
+    }
+
+    __syncthreads();
+
+    if (tile_idx * tile_size + (threadIdx.y * tx + threadIdx.x + 1) * vec_size - 1 < feat_in) {
+      y += sum;
+    }
+  }
+
+  if (threadIdx.x == 0) {
+    y_warpwise[threadIdx.y] = y;
+  }
+  __syncthreads();
+
+  float y_write = 0.f;
+#pragma unroll
+  for (size_t i = 0; i < ty; ++i) {
+    y_write += y_warpwise[i];
+  }
+ 
+  // write Y;
+  if (threadIdx.x == 0 && threadIdx.y == 0) {
+    size_t y_idx = batch_idx * full_y_size + y_offset + j;
+    Y[y_idx] = vllm_add<out_T>(Y[y_idx], convert_type<float, out_T>(y_write));
+  }
+}
+
+#endif
+
 // nthrs = (2, 16, 4)
 template <int feat_in, int feat_out, size_t vec_size, int tx, int ty, int tz,
           typename in_T, typename out_T, typename W_T>
@@ -172,7 +271,11 @@ bgmv_expand_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
   float sum = 0.f;
 #pragma unroll
   for (size_t i = 0; i < vec_size; ++i) {
+#ifndef USE_ROCM
     sum += float(w_vec[i]) * float(x_vec[i]) * scale;
+#else
+    sum += convert_type<W_T, float>(w_vec[i]) * convert_type<in_T, float>(x_vec[i]) * scale;
+#endif
   }
 
   cg::thread_block_tile g = cg::tiled_partition<tx>(block);
@@ -183,8 +286,14 @@ bgmv_expand_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
   sum = g.shfl(sum, 0);
 
   if (threadIdx.x == 0) {
+#ifndef USE_ROCM
     Y[batch_idx * full_y_size + y_offset + tile_idx * (tz * ty) +
       threadIdx.z * ty + threadIdx.y] += static_cast<out_T>(sum);
+#else
+    size_t y_idx = batch_idx * full_y_size + y_offset + tile_idx * (tz * ty) +
+                   threadIdx.z * ty + threadIdx.y;
+    Y[y_idx] = vllm_add<out_T>(Y[y_idx], convert_type<float, out_T>(sum));
+#endif
   }
 }
 
@@ -199,7 +308,7 @@ void bgmv_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
   constexpr int tz = 4;
   const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
 
-  if constexpr (feat_in < feat_out) {
+  if constexpr (feat_in <= feat_out) {
     static_assert(feat_in % vec_size == 0);
     constexpr int tx = feat_in / vec_size;
 
@@ -236,6 +345,7 @@ void bgmv_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
                                         scale);
     }
   } else {
+#ifndef USE_ROCM
     static_assert(feat_in % (vec_size * 32) == 0 ||
                   feat_in % (vec_size * 16) == 0 ||
                   feat_in % (vec_size * 8) == 0);
@@ -279,6 +389,50 @@ void bgmv_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
                                         full_y_size, num_layers, layer_idx,
                                         scale);
     }
+#else
+    constexpr size_t rocm_warp_size = warpSize;
+
+#define CHECK_INPUT_TILEABLE_BY(vec_size_) \
+    feat_in % (rocm_warp_size * vec_size_) == 0
+
+#define LAUNCH_BGMV_SHRINK_KERNELS_ROCM(factor_, vec_size_, tx_, ty_)       \
+    if constexpr (CHECK_INPUT_TILEABLE_BY(factor_)) {                       \
+      constexpr size_t vec_size_shrink = vec_size_;                         \
+      constexpr int tx = tx_;                                               \
+      constexpr int ty = ty_;                                               \
+      dim3 nblks(feat_out, batch_size);                                     \
+      dim3 nthrs(tx, ty);                                                   \
+      bgmv_shrink_kernel<feat_in, feat_out, vec_size_shrink,                \
+                          vec_size_shrink * sizeof(in_T),                   \
+                          vec_size_shrink * sizeof(W_T),                    \
+                          tx, ty, tz>                                       \
+          <<<nblks, nthrs, 0, stream>>>(Y, X, W, indicies, y_offset,        \
+                                        full_y_size, num_layers, layer_idx, \
+                                        scale);                             \
+    }
+
+    static_assert(CHECK_INPUT_TILEABLE_BY(32) ||
+                  CHECK_INPUT_TILEABLE_BY(16) ||
+                  CHECK_INPUT_TILEABLE_BY( 8) ||
+                  CHECK_INPUT_TILEABLE_BY( 4) ||
+                  CHECK_INPUT_TILEABLE_BY( 2) ||
+                  CHECK_INPUT_TILEABLE_BY( 1));
+    
+    LAUNCH_BGMV_SHRINK_KERNELS_ROCM(32, vec_size, rocm_warp_size, 32/vec_size)
+    else
+    LAUNCH_BGMV_SHRINK_KERNELS_ROCM(16, vec_size, rocm_warp_size, 16/vec_size)
+    else
+    LAUNCH_BGMV_SHRINK_KERNELS_ROCM( 8, vec_size, rocm_warp_size,  8/vec_size)
+    else
+    LAUNCH_BGMV_SHRINK_KERNELS_ROCM( 4, vec_size, rocm_warp_size/(vec_size/4), vec_size/4)
+    else
+    LAUNCH_BGMV_SHRINK_KERNELS_ROCM( 2, vec_size, rocm_warp_size/(vec_size/2), vec_size/2)
+    else
+    LAUNCH_BGMV_SHRINK_KERNELS_ROCM( 1, vec_size, rocm_warp_size/(vec_size/1), vec_size/1)
+
+#undef CHECK_INPUT_TILEABLE_BY
+#undef LAUNCH_BGMV_SHRINK_KERNELS_ROCM
+#endif
   }
 }
 
@@ -289,6 +443,9 @@ void bgmv_kernel(out_T *__restrict__ Y, const in_T *__restrict__ X,
       int64_t y_offset, int64_t full_y_size, int64_t batch_size,               \
       int64_t num_layers, int64_t layer_idx, float scale);
 
+#define INST_BGMV_ONESIDE(in_T, out_T, W_T, feat_in, feat_out)                 \
+  INST_BGMV(feat_in, feat_out, in_T, out_T, W_T)
+
 #define INST_BGMV_TWOSIDE(in_T, out_T, W_T, narrow, wide)                      \
   INST_BGMV(narrow, wide, in_T, out_T, W_T)                                    \
   INST_BGMV(wide, narrow, in_T, out_T, W_T)

csrc/punica/bgmv/generator.py

@@ -10,6 +10,7 @@ TEMPLATE = """
 #include "bgmv_impl.cuh"
 
 FOR_BGMV_WIDE_NARROW(INST_BGMV_TWOSIDE, {input_dtype}, {output_dtype}, {weight_dtype})
+FOR_INST_BGMV_WIDE_NARROW(INST_BGMV_ONESIDE, {input_dtype}, {output_dtype}, {weight_dtype})
 """.lstrip()  # noqa: E501
 
 for input_dtype in DTYPES:

csrc/punica/bgmv/vec_dtypes.cuh

@@ -1,8 +1,6 @@
 #ifndef VEC_DTYPES_CUH_
 #define VEC_DTYPES_CUH_
 
-#include <cuda_bf16.h>
-#include <cuda_fp16.h>
 #ifdef FLASHINFER_USE_FP8
 #include <cuda_fp8.h>
 #endif
@@ -10,6 +8,9 @@
 
 #include <type_traits>
 
+#include "../type_convert.h"
+#include "../../cuda_compat.h"
+
 #define FLASHINFER_INLINE \
   inline __attribute__((always_inline)) __device__ __host__
 

csrc/punica/punica_ops.cu

@@ -1,12 +1,11 @@
-#include <cuda_bf16.h>
+#include <torch/all.h>
-#include <cuda_fp16.h>
-#include <torch/extension.h>
 #include <c10/cuda/CUDAGuard.h>
 #include <cstdint>
 
+#include "type_convert.h"
+#include "../cuda_compat.h"
 #include "bgmv/bgmv_config.h"
 
-namespace {
 
 //====== utils ======
 
@@ -79,17 +78,17 @@ inline bool launch_bgmv_kernel(out_T *Y, const in_T *X, const W_T *W,
   CASE_ONESIDE(in_T, out_T, W_T, wide, narrow)
 
     FOR_BGMV_WIDE_NARROW(CASE, _, _, _)
+    FOR_INST_BGMV_WIDE_NARROW(CASE_ONESIDE, _, _, _)
 #undef CASE
 #undef CASE_ONESIDE
   default:
     return false;
   }
-
   return true;
 }
 
 void dispatch_bgmv(torch::Tensor y, torch::Tensor x, torch::Tensor w,
-                   torch::Tensor indicies, int64_t layer_idx, float scale) {
+                   torch::Tensor indicies, int64_t layer_idx, double scale) {
   CHECK_INPUT(y);
   CHECK_INPUT(x);
   CHECK_INPUT(w);
@@ -321,7 +320,7 @@ void dispatch_bgmv(torch::Tensor y, torch::Tensor x, torch::Tensor w,
 
 void dispatch_bgmv_low_level(torch::Tensor y, torch::Tensor x, torch::Tensor w,
                              torch::Tensor indicies, int64_t layer_idx,
-                             float scale, int64_t h_in, int64_t h_out,
+                             double scale, int64_t h_in, int64_t h_out,
                              int64_t y_offset) {
   CHECK_INPUT(y);
   CHECK_INPUT(x);
@@ -568,15 +567,3 @@ void dispatch_bgmv_low_level(torch::Tensor y, torch::Tensor x, torch::Tensor w,
   TORCH_CHECK(ok, "No suitable kernel.", " h_in=", h_in, " h_out=", h_out,
               " dtype=", x.scalar_type(), " out_dtype=", y.scalar_type());
 }
-
-} // namespace
-
-//====== pybind ======
-
-#define DEFINE_pybind(name) m.def(#name, &name, #name);
-
-PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
-  m.def("dispatch_bgmv", &dispatch_bgmv, "dispatch_bgmv");
-  m.def("dispatch_bgmv_low_level", &dispatch_bgmv_low_level,
-        "dispatch_bgmv_low_level");
-}

csrc/punica/punica_ops.h

@@ -0,0 +1,11 @@
+#pragma once
+
+#include <torch/all.h>
+
+void dispatch_bgmv(torch::Tensor y, torch::Tensor x, torch::Tensor w,
+                   torch::Tensor indicies, int64_t layer_idx, double scale);
+
+void dispatch_bgmv_low_level(torch::Tensor y, torch::Tensor x, torch::Tensor w,
+                             torch::Tensor indicies, int64_t layer_idx,
+                             double scale, int64_t h_in, int64_t h_out,
+                             int64_t y_offset);

csrc/punica/torch_bindings.cpp

@@ -0,0 +1,18 @@
+#include "registration.h"
+#include "punica_ops.h"
+
+TORCH_LIBRARY_EXPAND(TORCH_EXTENSION_NAME, m) {
+  m.def(
+      "dispatch_bgmv(Tensor! y, Tensor x, Tensor w, Tensor indicies, int "
+      "layer_idx, float scale) -> ()");
+  m.impl("dispatch_bgmv", torch::kCUDA, &dispatch_bgmv);
+
+  m.def(
+      "dispatch_bgmv_low_level(Tensor! y, Tensor x, Tensor w,"
+      "Tensor indicies, int layer_idx,"
+      "float scale, int h_in, int h_out,"
+      "int y_offset) -> ()");
+  m.impl("dispatch_bgmv_low_level", torch::kCUDA, &dispatch_bgmv_low_level);
+}
+
+REGISTER_EXTENSION(TORCH_EXTENSION_NAME)

csrc/punica/type_convert.h

@@ -0,0 +1,82 @@
+#ifndef CSRC__PUNICA__TYPE_CONVERT_H__
+#define CSRC__PUNICA__TYPE_CONVERT_H__
+
+#ifndef USE_ROCM
+
+#include <cuda_bf16.h>
+#include <cuda_fp16.h>
+
+#else
+
+#include <hip/hip_bf16.h>
+#include <hip/hip_fp16.h>
+
+#define __TYPE_CONVERT__HOST_DEVICE__ __host__ __device__
+
+typedef __half nv_half;
+typedef __hip_bfloat16 nv_bfloat16;
+typedef __hip_bfloat162 nv_bfloat162;
+
+__TYPE_CONVERT__HOST_DEVICE__
+inline __hip_bfloat162 make_bfloat162(__hip_bfloat16 val) {
+  return __hip_bfloat162{val, val};
+}
+
+__TYPE_CONVERT__HOST_DEVICE__
+inline __hip_bfloat162 make_bfloat162(__hip_bfloat16 vall, __hip_bfloat16 valr) {
+  return __hip_bfloat162{vall, valr};
+}
+
+template <typename T_src, typename T_dst>
+__TYPE_CONVERT__HOST_DEVICE__
+inline T_dst convert_type(T_src val) {
+  return static_cast<T_dst>(val);
+}
+
+template <>
+__TYPE_CONVERT__HOST_DEVICE__
+inline float convert_type<__half, float>(__half val) {
+  return __half2float(val);
+}
+
+template <>
+__TYPE_CONVERT__HOST_DEVICE__
+inline __half convert_type<float, __half>(float val) {
+  return __float2half(val);
+}
+
+template <>
+__TYPE_CONVERT__HOST_DEVICE__
+inline float convert_type<__hip_bfloat16, float>(__hip_bfloat16 val) {
+  return __bfloat162float(val);
+}
+
+template <>
+__TYPE_CONVERT__HOST_DEVICE__
+inline __hip_bfloat16 convert_type<float, __hip_bfloat16>(float val) {
+  return __float2bfloat16(val);
+}
+
+template <typename T>
+__TYPE_CONVERT__HOST_DEVICE__
+inline T vllm_add(T a, T b) {
+  return a + b;
+}
+
+template <>
+__TYPE_CONVERT__HOST_DEVICE__
+inline __half vllm_add<__half>(__half a, __half b) {
+  return __hadd(a, b);
+}
+
+template <>
+__TYPE_CONVERT__HOST_DEVICE__
+inline __hip_bfloat16 vllm_add<__hip_bfloat16>(__hip_bfloat16 a, __hip_bfloat16 b) {
+  return __hadd(a, b);
+}
+
+#undef __TYPE_CONVERT__HOST_DEVICE__
+
+#endif // USE_ROCM
+
+#endif // CSRC__PUNICA__TYPE_CONVERT_H__

csrc/pybind.cpp

@@ -1,129 +0,0 @@
-#include "cache.h"
-#include "cuda_utils.h"
-#include "ops.h"
-#include <torch/extension.h>
-
-PYBIND11_MODULE(TORCH_EXTENSION_NAME, m) {
-  // vLLM custom ops
-  pybind11::module ops = m.def_submodule("ops", "vLLM custom operators");
-
-  // Attention ops
-  ops.def(
-    "paged_attention_v1",
-    &paged_attention_v1,
-    "Compute the attention between an input query and the cached keys/values using PagedAttention.");
-  ops.def(
-    "paged_attention_v2",
-    &paged_attention_v2,
-    "PagedAttention V2.");
-
-  // Activation ops
-  ops.def(
-    "silu_and_mul",
-    &silu_and_mul,
-    "Activation function used in SwiGLU.");
-  ops.def(
-    "gelu_and_mul",
-    &gelu_and_mul,
-    "Activation function used in GeGLU with `none` approximation.");
-  ops.def(
-    "gelu_tanh_and_mul",
-    &gelu_tanh_and_mul,
-    "Activation function used in GeGLU with `tanh` approximation.");
-  ops.def(
-    "gelu_new",
-    &gelu_new,
-    "GELU implementation used in GPT-2.");
-  ops.def(
-    "gelu_fast",
-    &gelu_fast,
-    "Approximate GELU implementation.");
-
-  // Layernorm
-  ops.def(
-    "rms_norm",
-    &rms_norm,
-    "Apply Root Mean Square (RMS) Normalization to the input tensor.");
-
-  ops.def(
-    "fused_add_rms_norm",
-    &fused_add_rms_norm,
-    "In-place fused Add and RMS Normalization");
-
-  // Rotary embedding
-  ops.def(
-    "rotary_embedding",
-    &rotary_embedding,
-    "Apply GPT-NeoX or GPT-J style rotary embedding to query and key");
-
-  ops.def(
-    "batched_rotary_embedding",
-    &batched_rotary_embedding,
-    "Apply GPT-NeoX or GPT-J style rotary embedding to query and key (supports multiple loras)");
-
-// Quantization ops
-#ifndef USE_ROCM
-  ops.def("aqlm_gemm", &aqlm_gemm, "Quantized GEMM for AQLM");
-  ops.def("aqlm_dequant", &aqlm_dequant, "Decompression method for AQLM");
-  ops.def("awq_gemm", &awq_gemm, "Quantized GEMM for AWQ");
-  ops.def("marlin_gemm", &marlin_gemm, "Marlin Optimized Quantized GEMM for GPTQ");
-  ops.def("awq_dequantize", &awq_dequantize, "Dequantization for AWQ");
-#endif
- 
-  ops.def("gptq_gemm", &gptq_gemm, "Quantized GEMM for GPTQ");
-  ops.def("gptq_shuffle", &gptq_shuffle, "Post processing for GPTQ");
-  ops.def("squeezellm_gemm", &squeezellm_gemm, "Quantized GEMM for SqueezeLLM");
-  ops.def("scaled_fp8_quant", &scaled_fp8_quant, "Compute FP8 quantized tensor and scaling factor");
-  ops.def(
-    "moe_align_block_size",
-    &moe_align_block_size,
-    "Aligning the number of tokens to be processed by each expert such that it is divisible by the block size.");
-
-  // Cache ops
-  pybind11::module cache_ops = m.def_submodule("cache_ops", "vLLM cache ops");
-  cache_ops.def(
-    "swap_blocks",
-    &swap_blocks,
-    "Swap in (out) the cache blocks from src to dst");
-  cache_ops.def(
-    "copy_blocks",
-    &copy_blocks,
-    "Copy the cache blocks from src to dst");
-  cache_ops.def(
-    "reshape_and_cache",
-    &reshape_and_cache,
-    "Reshape the key and value tensors and cache them");
-  cache_ops.def(
-    "convert_fp8",
-    &convert_fp8,
-    "Convert the key and value cache to fp8 data type");
-
-  // Cuda utils
-  pybind11::module cuda_utils = m.def_submodule("cuda_utils", "vLLM cuda utils");
-  cuda_utils.def(
-    "get_device_attribute",
-    &get_device_attribute,
-    "Gets the specified device attribute.");
-
-  cuda_utils.def(
-    "get_max_shared_memory_per_block_device_attribute",
-    &get_max_shared_memory_per_block_device_attribute,
-    "Gets the maximum shared memory per block device attribute.");
-
-#ifndef USE_ROCM
-  // Custom all-reduce kernels
-  pybind11::module custom_ar = m.def_submodule("custom_ar", "custom allreduce");
-  custom_ar.def("init_custom_ar", &init_custom_ar, "init_custom_ar");
-  custom_ar.def("should_custom_ar", &should_custom_ar, "should_custom_ar");
-  custom_ar.def("all_reduce_reg", &all_reduce_reg, "all_reduce_reg");
-  custom_ar.def("all_reduce_unreg", &all_reduce_unreg, "all_reduce_unreg");
-  custom_ar.def("dispose", &dispose, "dispose");
-  custom_ar.def("meta_size", &meta_size, "meta_size");
-  custom_ar.def("register_buffer", &register_buffer, "register_buffer");
-  custom_ar.def("get_graph_buffer_ipc_meta", &get_graph_buffer_ipc_meta,
-                "get_graph_buffer_ipc_meta");
-  custom_ar.def("register_graph_buffers", &register_graph_buffers,
-                "register_graph_buffers");
-#endif
-
-}

csrc/quantization/aqlm/gemm_kernels.cu

@@ -18,37 +18,33 @@
 #include <cuda.h>
 #include <cuda_fp16.h>
 #include <cuda_runtime.h>
-#include <torch/extension.h>
+#include <torch/all.h>
 #include <c10/cuda/CUDAStream.h>
 #include <c10/cuda/CUDAGuard.h>
 
 #include <iostream>
 #include <cstdlib>
 
-
 namespace vllm {
 namespace aqlm {
 
 __global__ void Code1x16MatVec(
-  const int4* __restrict__ A,
+    const int4* __restrict__ A, const int4* __restrict__ B,
-  const int4* __restrict__ B,
+    int4* __restrict__ C, const int4* __restrict__ codebook, const int prob_m,
-        int4* __restrict__ C,
-  const int4* __restrict__ codebook,
-  const int prob_m,
     const int prob_k,
-  const int4 codebook_a_sizes,  // cumulative sizes of A spanning each codebook, at most 3 long.
+    const int4 codebook_a_sizes,  // cumulative sizes of A spanning each
+                                  // codebook, at most 3 long.
     const int codebook_stride     // as int4.
 ) {
   int a_gl_stride = prob_k / 8 / 8;
   int a_gl_rd = (blockDim.x / 32) * blockIdx.x + (threadIdx.x / 32);
   bool pred = a_gl_rd < prob_m;
 
-  if (pred)
+  if (pred) {
-  {
+    // advance to the correct codebook, this easy because we only multiply one
-    // advance to the correct codebook, this easy because we only multiply one column of the codebook.
+    // column of the codebook.
     auto codebook_size = &codebook_a_sizes.x;
-    while (a_gl_rd >= *codebook_size)
+    while (a_gl_rd >= *codebook_size) {
-    {
       codebook += codebook_stride;
       ++codebook_size;
     }
@@ -67,8 +63,7 @@ __global__ void Code1x16MatVec(
     // We pad shared memory to avoid bank conflicts during reads
     __syncthreads();
     for (int i = threadIdx.x; i < 32 * 8; i += blockDim.x) {
-      if (b_gl_rd + i < prob_k / 8)
+      if (b_gl_rd + i < prob_k / 8) sh_b[9 * (i / 8) + i % 8] = B[b_gl_rd + i];
-        sh_b[9 * (i / 8) + i % 8] = B[b_gl_rd + i];
     }
     __syncthreads();
     b_gl_rd += 32 * 8;
@@ -79,19 +74,16 @@ __global__ void Code1x16MatVec(
 #pragma unroll
       for (int i = 0; i < 8; i++) {
         uint32_t dec[4];
-        // We bypass the L1 cache to avoid massive amounts of memory streaming that doesn't
+        // We bypass the L1 cache to avoid massive amounts of memory streaming
-        // actually help us; this brings > 2x speedup.
+        // that doesn't actually help us; this brings > 2x speedup.
-        asm volatile (
+        asm volatile("ld.cg.global.v4.u32 {%0, %1, %2, %3}, [%4];"
-          "ld.cg.global.v4.u32 {%0, %1, %2, %3}, [%4];"
                      : "=r"(dec[0]), "=r"(dec[1]), "=r"(dec[2]), "=r"(dec[3])
-          : "l"((void*) &codebook[enc[i]])
+                     : "l"((void*)&codebook[enc[i]]));
-        );
         half2* a = reinterpret_cast<half2*>(&dec);
         half2* b = reinterpret_cast<half2*>(&sh_b[b_sh_rd]);
         half2 res2 = {};
 #pragma unroll
-        for (int j = 0; j < 4; j++)
+        for (int j = 0; j < 4; j++) res2 = __hfma2(a[j], b[j], res2);
-          res2 = __hfma2(a[j], b[j], res2);
         res += __half2float(res2.x) + __half2float(res2.y);
         b_sh_rd++;
       }
@@ -101,21 +93,18 @@ __global__ void Code1x16MatVec(
 
   if (pred) {
 #pragma unroll
-    for (int i = 16; i > 0; i /= 2)
+    for (int i = 16; i > 0; i /= 2) res += __shfl_down_sync(0xffffffff, res, i);
-      res += __shfl_down_sync(0xffffffff, res, i);
     if (threadIdx.x % 32 == 0)
       reinterpret_cast<__half*>(C)[c_gl_wr] = __float2half(res);
   }
 }
 
 __global__ void Code2x8MatVec(
-  const int4* __restrict__ A,
+    const int4* __restrict__ A, const int4* __restrict__ B,
-  const int4* __restrict__ B,
+    int4* __restrict__ C, const int4* __restrict__ codebook, int prob_m,
-        int4* __restrict__ C,
-  const int4* __restrict__ codebook,
-  int prob_m,
     int prob_k,
-  const int4 codebook_a_sizes,  // cumulative sizes of A spanning each codebook, at most 3 long.
+    const int4 codebook_a_sizes,  // cumulative sizes of A spanning each
+                                  // codebook, at most 3 long.
     const int codebook_stride     // as int4.
 
 ) {
@@ -123,12 +112,11 @@ __global__ void Code2x8MatVec(
   int a_gl_rd = (blockDim.x / 32) * blockIdx.x + (threadIdx.x / 32);
   bool pred = a_gl_rd < prob_m;
 
-  if (pred)
+  if (pred) {
-  {
+    // advance to the correct codebook, this easy because we only multiply one
-    // advance to the correct codebook, this easy because we only multiply one column of the codebook.
+    // column of the codebook.
     auto codebook_size = &codebook_a_sizes.x;
-    while (a_gl_rd >= *codebook_size)
+    while (a_gl_rd >= *codebook_size) {
-    {
       codebook += codebook_stride;
       ++codebook_size;
     }
@@ -149,8 +137,7 @@ __global__ void Code2x8MatVec(
   for (int i = threadIdx.x; i < 2 * 256; i += blockDim.x) {
     int4 dec = codebook[i];
 #pragma unroll
-    for (int j = 0; j < 8; j++)
+    for (int j = 0; j < 8; j++) sh_code[8 * i + (j + lane) % 8] = dec;
-      sh_code[8 * i + (j + lane) % 8] = dec;
   }
   __syncthreads();
 
@@ -161,8 +148,7 @@ __global__ void Code2x8MatVec(
     // We pad shared memory to avoid bank conflicts during reads
     __syncthreads();
     for (int i = threadIdx.x; i < 32 * 8; i += blockDim.x) {
-      if (b_gl_rd + i < prob_k / 8)
+      if (b_gl_rd + i < prob_k / 8) sh_b[9 * (i / 8) + i % 8] = B[b_gl_rd + i];
-        sh_b[9 * (i / 8) + i % 8] = B[b_gl_rd + i];
     }
     __syncthreads();
     b_gl_rd += 32 * 8;
@@ -172,8 +158,10 @@ __global__ void Code2x8MatVec(
       const uint8_t* enc = reinterpret_cast<const uint8_t*>(&A[a_gl_rd]);
 #pragma unroll
       for (int i = 0; i < 8; i++) {
-        half2* a0 = reinterpret_cast<half2*>(&sh_code0[8 * enc[2 * i + 0] + lane]);
+        half2* a0 =
-        half2* a1 = reinterpret_cast<half2*>(&sh_code1[8 * enc[2 * i + 1] + lane]);
+            reinterpret_cast<half2*>(&sh_code0[8 * enc[2 * i + 0] + lane]);
+        half2* a1 =
+            reinterpret_cast<half2*>(&sh_code1[8 * enc[2 * i + 1] + lane]);
         half2* b = reinterpret_cast<half2*>(&sh_b[b_sh_rd]);
         half2 res2 = {};
 #pragma unroll
@@ -188,33 +176,28 @@ __global__ void Code2x8MatVec(
 
   if (pred) {
 #pragma unroll
-    for (int i = 16; i > 0; i /= 2)
+    for (int i = 16; i > 0; i /= 2) res += __shfl_down_sync(0xffffffff, res, i);
-      res += __shfl_down_sync(0xffffffff, res, i);
     if (threadIdx.x % 32 == 0)
       reinterpret_cast<__half*>(C)[c_gl_wr] = __float2half(res);
   }
 }
 
-
 __global__ void Code1x16Dequant(
-  const int4* __restrict__ A,
+    const int4* __restrict__ A, int4* __restrict__ C,
-        int4* __restrict__ C,
+    const int4* __restrict__ codebook, int prob_m, int prob_k,
-  const int4* __restrict__ codebook,
+    const int4 codebook_a_sizes,  // cumulative sizes of A spanning each
-  int prob_m,
+                                  // codebook, at most 3 long, sums to m.
-  int prob_k,
-  const int4 codebook_a_sizes,  // cumulative sizes of A spanning each codebook, at most 3 long, sums to m.
     const int codebook_stride     // as int4
 ) {
   int a_gl_stride = prob_k / 8 / 8;
   int a_gl_rd = (blockDim.x / 32) * blockIdx.x + (threadIdx.x / 32);
   bool pred = a_gl_rd < prob_m;
 
-  if (pred)
+  if (pred) {
-  {
+    // advance to the correct codebook, this easy because we only multiply one
-    // advance to the correct codebook, this easy because we only multiply one column of the codebook.
+    // column of the codebook.
     auto codebook_size = &codebook_a_sizes.x;
-    while (a_gl_rd >= *codebook_size)
+    while (a_gl_rd >= *codebook_size) {
-    {
       codebook += codebook_stride;
       ++codebook_size;
     }
@@ -235,13 +218,11 @@ __global__ void Code1x16Dequant(
       for (int i = 0; i < 8; i++) {
         int4 chunk;
         auto dec = reinterpret_cast<uint32_t*>(&chunk);
-        // We bypass the L1 cache to avoid massive amounts of memory streaming that doesn't
+        // We bypass the L1 cache to avoid massive amounts of memory streaming
-        // actually help us; this brings > 2x speedup.
+        // that doesn't actually help us; this brings > 2x speedup.
-        asm volatile (
+        asm volatile("ld.cg.global.v4.u32 {%0, %1, %2, %3}, [%4];"
-          "ld.cg.global.v4.u32 {%0, %1, %2, %3}, [%4];"
                      : "=r"(dec[0]), "=r"(dec[1]), "=r"(dec[2]), "=r"(dec[3])
-          : "l"((void*) &codebook[enc[i]])
+                     : "l"((void*)&codebook[enc[i]]));
-        );
 
         C[a_gl_rd * 8 + i] = chunk;
       }
@@ -250,26 +231,23 @@ __global__ void Code1x16Dequant(
   }
 }
 
-
 __global__ void Code2x8Dequant(
-  const int4* __restrict__ A,
+    const int4* __restrict__ A, int4* __restrict__ C,
-        int4* __restrict__ C,
+    const int4* __restrict__ codebook, int prob_m, int prob_k,
-  const int4* __restrict__ codebook,
+    const int4
-  int prob_m,
+        codebook_a_sizes,  // cumulative sizes of A spanning each codebook, at
-  int prob_k,
+                           // most 3 long, corresponds to cols.
-  const int4 codebook_a_sizes,  // cumulative sizes of A spanning each codebook, at most 3 long, corresponds to cols.
     const int codebook_stride  // as int4
 ) {
   int a_gl_stride = prob_k / 8 / 8;
   int a_gl_rd = (blockDim.x / 32) * blockIdx.x + (threadIdx.x / 32);
   bool pred = a_gl_rd < prob_m;
 
-  if (pred)
+  if (pred) {
-  {
+    // advance to the correct codebook, this easy because we only multiply one
-    // advance to the correct codebook, this easy because we only multiply one column of the codebook.
+    // column of the codebook.
     auto codebook_size = &codebook_a_sizes.x;
-    while (a_gl_rd >= *codebook_size)
+    while (a_gl_rd >= *codebook_size) {
-    {
       codebook += codebook_stride;
       ++codebook_size;
     }
@@ -291,8 +269,7 @@ __global__ void Code2x8Dequant(
   for (int i = threadIdx.x; i < 2 * 256; i += blockDim.x) {
     int4 dec = codebook[i];
 #pragma unroll
-    for (int j = 0; j < 8; j++)
+    for (int j = 0; j < 8; j++) sh_code[8 * i + (j + lane) % 8] = dec;
-      sh_code[8 * i + (j + lane) % 8] = dec;
   }
   __syncthreads();
 
@@ -305,8 +282,10 @@ __global__ void Code2x8Dequant(
 #pragma unroll
       for (int i = 0; i < 8; i++) {
         int4 chunk;
-        half2* a0 = reinterpret_cast<half2*>(&sh_code0[8 * enc[2 * i + 0] + lane]);
+        half2* a0 =
-        half2* a1 = reinterpret_cast<half2*>(&sh_code1[8 * enc[2 * i + 1] + lane]);
+            reinterpret_cast<half2*>(&sh_code0[8 * enc[2 * i + 0] + lane]);
+        half2* a1 =
+            reinterpret_cast<half2*>(&sh_code1[8 * enc[2 * i + 1] + lane]);
 #pragma unroll
         for (int j = 0; j < 4; j++)
           reinterpret_cast<half2*>(&chunk)[j] = __hadd2(a0[j], a1[j]);
@@ -317,22 +296,15 @@ __global__ void Code2x8Dequant(
   }
 }
 
-inline int ceildiv(int a, int b) {
+inline int ceildiv(int a, int b) { return (a + b - 1) / b; }
-  return (a + b - 1) / b;
-}
 
 const int THREAD_M = 16;
 
-void  code1x16_matvec_cuda(
+void code1x16_matvec_cuda(const void* __restrict__ A,
-  const void* __restrict__ A,
+                          const void* __restrict__ B, void* __restrict__ C,
-  const void* __restrict__ B,
+                          const void* __restrict__ codebook, int prob_m,
-        void* __restrict__ C,
+                          int prob_k, const int4 codebook_a_sizes,
-  const void* __restrict__ codebook,
+                          const int codebook_stride) {
-  int prob_m,
-  int prob_k,
-  const int4 codebook_a_sizes,
-  const int codebook_stride
-) {
   int sms;
   cudaDeviceGetAttribute(&sms, cudaDevAttrMultiProcessorCount, 0);
   int waves = 0;
@@ -346,27 +318,15 @@ void  code1x16_matvec_cuda(
   int threads = 32 * thread_m;
   cudaStream_t stream = at::cuda::getCurrentCUDAStream().stream();
   Code1x16MatVec<<<blocks, threads, 16 * 32 * 9, stream>>>(
-    (const int4*) A,
+      (const int4*)A, (const int4*)B, (int4*)C, (const int4*)codebook, prob_m,
-    (const int4*) B,
+      prob_k, codebook_a_sizes, codebook_stride);
-    (int4*) C,
-    (const int4*) codebook,
-    prob_m,
-    prob_k,
-    codebook_a_sizes,
-    codebook_stride
-  );
 }
 
-void  code2x8_matvec_cuda(
+void code2x8_matvec_cuda(const void* __restrict__ A, const void* __restrict__ B,
-  const void* __restrict__ A,
-  const void* __restrict__ B,
                          void* __restrict__ C,
-  const void* __restrict__ codebook,
+                         const void* __restrict__ codebook, int prob_m,
-  int prob_m,
+                         int prob_k, const int4 codebook_a_sizes,
-  int prob_k,
+                         const int codebook_stride) {
-  const int4 codebook_a_sizes,
-  const int codebook_stride
-) {
   int sms;
   cudaDeviceGetAttribute(&sms, cudaDevAttrMultiProcessorCount, 0);
   int waves = 0;
@@ -379,29 +339,19 @@ void  code2x8_matvec_cuda(
   int blocks = ceildiv(prob_m, thread_m);
   int threads = 32 * thread_m;
   int shared = 16 * (2 * 256 * 8 + 32 * 9);
-  cudaFuncSetAttribute(
+  cudaFuncSetAttribute(Code2x8MatVec,
-    Code2x8MatVec, cudaFuncAttributeMaxDynamicSharedMemorySize, shared
+                       cudaFuncAttributeMaxDynamicSharedMemorySize, shared);
-  );
   cudaStream_t stream = at::cuda::getCurrentCUDAStream().stream();
   Code2x8MatVec<<<blocks, threads, shared, stream>>>(
-    (const int4*) A,
+      (const int4*)A, (const int4*)B, (int4*)C, (const int4*)codebook, prob_m,
-    (const int4*) B,
+      prob_k, codebook_a_sizes, codebook_stride);
-    (int4*) C,
-    (const int4*) codebook,
-    prob_m,
-    prob_k,
-    codebook_a_sizes,
-    codebook_stride
-  );
 }
 
 void code1x16_dequant_cuda(
-  const void* __restrict__ A,
+    const void* __restrict__ A, void* __restrict__ C,
-        void* __restrict__ C,
+    const void* __restrict__ codebook, int prob_m, int prob_k,
-  const void* __restrict__ codebook,
+    const int4 codebook_a_sizes,  // cumulative sizes of A spanning each
-  int prob_m,
+                                  // codebook, at most 3 long.
-  int prob_k,
-  const int4 codebook_a_sizes,  // cumulative sizes of A spanning each codebook, at most 3 long.
     const int codebook_stride     // as int4.
 ) {
   int sms;
@@ -417,24 +367,20 @@ void code1x16_dequant_cuda(
   int threads = 32 * thread_m;
   cudaStream_t stream = at::cuda::getCurrentCUDAStream().stream();
   Code1x16Dequant<<<blocks, threads, 0, stream>>>(
-    (const int4*) A,
+      (const int4*)A, (int4*)C, (const int4*)codebook, prob_m, prob_k,
-    (int4*) C,
+      codebook_a_sizes,  // cumulative sizes of A spanning each codebook, at
-    (const int4*) codebook,
+                         // most 3 long.
-    prob_m,
-    prob_k,
-    codebook_a_sizes,  // cumulative sizes of A spanning each codebook, at most 3 long.
       codebook_stride    // as int4.
   );
 }
 
 // Dequantizes the code and codebook into weights.
 void code2x8_dequant_cuda(
-  const void* __restrict__ A,
+    const void* __restrict__ A, void* __restrict__ C,
-        void* __restrict__ C,
+    const void* __restrict__ codebook, int prob_m, int prob_k,
-  const void* __restrict__ codebook,
+    const int4
-  int prob_m,
+        codebook_a_sizes,  // cumulative sizes of A spanning each codebook, at
-  int prob_k,
+                           // most 3 long, corresponds to cols.
-  const int4 codebook_a_sizes,  // cumulative sizes of A spanning each codebook, at most 3 long, corresponds to cols.
     const int codebook_stride  // as int4
 ) {
   int sms;
@@ -451,50 +397,33 @@ void  code2x8_dequant_cuda(
   int shared = 16 * (2 * 256 * 8 + 32 * 9);
   cudaStream_t stream = at::cuda::getCurrentCUDAStream().stream();
 
-  cudaFuncSetAttribute(
+  cudaFuncSetAttribute(Code2x8Dequant,
-    Code2x8Dequant, cudaFuncAttributeMaxDynamicSharedMemorySize, shared
+                       cudaFuncAttributeMaxDynamicSharedMemorySize, shared);
-  );
   Code2x8Dequant<<<blocks, threads, shared, stream>>>(
-    (const int4*) A,
+      (const int4*)A, (int4*)C, (const int4*)codebook, prob_m, prob_k,
-    (int4*) C,
+      codebook_a_sizes, codebook_stride);
-    (const int4*) codebook,
-    prob_m,
-    prob_k,
-    codebook_a_sizes,
-    codebook_stride
-  );
 }
 
-int codebook_stride(const torch::Tensor& codebooks)
+int codebook_stride(const torch::Tensor& codebooks) {
-{
   return codebooks.stride(0) * codebooks.element_size() / sizeof(int4);
 }
 
 void code1x16_matvec(
-  const torch::Tensor& A,
+    const torch::Tensor& A, const torch::Tensor& B, torch::Tensor& C,
-  const torch::Tensor& B,
-        torch::Tensor& C,
     const torch::Tensor& codebook,
-  const int4 codebook_a_sizes  // cumulative sizes of A spanning each codebook, at most 3 long.
+    const int4 codebook_a_sizes  // cumulative sizes of A spanning each
+                                 // codebook, at most 3 long.
 ) {
   const at::cuda::OptionalCUDAGuard device_guard(device_of(A));
   int prob_m = C.size(0);
   int prob_k = B.size(0);
 
-  code1x16_matvec_cuda(
+  code1x16_matvec_cuda(A.data_ptr(), B.data_ptr(), C.data_ptr(),
-    A.data_ptr(),
+                       codebook.data_ptr(), prob_m, prob_k, codebook_a_sizes,
-    B.data_ptr(),
+                       codebook_stride(codebook));
-    C.data_ptr(),
-    codebook.data_ptr(),
-    prob_m,
-    prob_k,
-    codebook_a_sizes,
-    codebook_stride(codebook)
-  );
 }
 
-torch::Tensor code1x16_matmat(
+torch::Tensor code1x16_matmat(const torch::Tensor& input,
-  const torch::Tensor& input,
                               const torch::Tensor& codes,
                               const torch::Tensor& codebooks,
                               const torch::Tensor& scales,
@@ -503,22 +432,15 @@ torch::Tensor code1x16_matmat(
   auto input_sizes = input.sizes();
   auto out_features = codes.size(0) * codebooks.size(2);
   auto flat_input = input.reshape({-1, input.size(-1)});
-  auto flat_output = torch::empty({flat_input.size(0), out_features},
+  auto flat_output = torch::empty(
-    torch::TensorOptions()
+      {flat_input.size(0), out_features},
-      .dtype(input.dtype())
+      torch::TensorOptions().dtype(input.dtype()).device(input.device()));
-      .device(input.device())
-  );
 
   for (int i = 0; i < flat_input.size(0); ++i) {
     auto input_vec = flat_input.index({i});
     auto output_vec = flat_output.index({i});
-    code1x16_matvec(
+    code1x16_matvec(codes.squeeze(2), input_vec, output_vec, codebooks,
-      codes.squeeze(2),
+                    codebook_a_sizes);
-      input_vec,
-      output_vec,
-      codebooks,
-      codebook_a_sizes
-    );
   }
   flat_output *= scales.flatten().unsqueeze(0);
 
@@ -533,55 +455,35 @@ torch::Tensor code1x16_matmat(
   return output;
 }
 
-void code2x8_matvec(
+void code2x8_matvec(const torch::Tensor& A, const torch::Tensor& B,
-  const torch::Tensor& A,
+                    torch::Tensor& C, const torch::Tensor& codebook,
-  const torch::Tensor& B,
+                    const int4 codebook_a_sizes) {
-        torch::Tensor& C,
-  const torch::Tensor& codebook,
-  const int4 codebook_a_sizes
-) {
   const at::cuda::OptionalCUDAGuard device_guard(device_of(A));
   int prob_m = C.size(0);
   int prob_k = B.size(0);
-  code2x8_matvec_cuda(
+  code2x8_matvec_cuda(A.data_ptr(), B.data_ptr(), C.data_ptr(),
-    A.data_ptr(),
+                      codebook.data_ptr(), prob_m, prob_k, codebook_a_sizes,
-    B.data_ptr(),
+                      2 * codebook_stride(codebook));
-    C.data_ptr(),
-    codebook.data_ptr(),
-    prob_m,
-    prob_k,
-    codebook_a_sizes,
-    2 * codebook_stride(codebook)
-  );
 }
 
-torch::Tensor code2x8_matmat(
+torch::Tensor code2x8_matmat(const torch::Tensor& input,
-  const torch::Tensor& input,
                              const torch::Tensor& codes,
                              const torch::Tensor& codebooks,
                              const torch::Tensor& scales,
                              const int4 codebook_a_sizes,
-  const std::optional<torch::Tensor>& bias
+                             const std::optional<torch::Tensor>& bias) {
-) {
   auto input_sizes = input.sizes();
   auto out_features = codes.size(0) * codebooks.size(2);
   auto flat_input = input.reshape({-1, input.size(-1)});
-  auto flat_output = torch::empty({flat_input.size(0), out_features},
+  auto flat_output = torch::empty(
-    torch::TensorOptions()
+      {flat_input.size(0), out_features},
-      .dtype(input.dtype())
+      torch::TensorOptions().dtype(input.dtype()).device(input.device()));
-      .device(input.device())
-  );
 
   for (int i = 0; i < flat_input.size(0); ++i) {
     auto input_vec = flat_input.index({i});
     auto output_vec = flat_output.index({i});
-    code2x8_matvec(
+    code2x8_matvec(codes.squeeze(2), input_vec, output_vec, codebooks,
-      codes.squeeze(2),
+                   codebook_a_sizes);
-      input_vec,
-      output_vec,
-      codebooks,
-      codebook_a_sizes
-    );
   }
   flat_output *= scales.flatten().unsqueeze(0);
   if (bias.has_value()) {
@@ -596,21 +498,18 @@ torch::Tensor code2x8_matmat(
 }
 
 // Accumulate the partition sizes.
-int4 accumulate_sizes(const torch::Tensor& codebook_partition_sizes)
+int4 accumulate_sizes(const torch::Tensor& codebook_partition_sizes) {
-{
   int4 cumulative_sizes;
   auto cumulative_size = &cumulative_sizes.x;
   int i = 0;
   int last = 0;
   assert(codebook_partition_sizes.size(0) <= 4);
-  for (; i <  codebook_partition_sizes.size(0); ++i, ++cumulative_size)
+  for (; i < codebook_partition_sizes.size(0); ++i, ++cumulative_size) {
-  {
     *cumulative_size = codebook_partition_sizes[i].item<int>() + last;
     last = *cumulative_size;
   }
   // fill in the rest with unreachable.
-  for (; i < 4; ++i, ++cumulative_size)
+  for (; i < 4; ++i, ++cumulative_size) {
-  {
     *cumulative_size = last * 10;
   }
   return cumulative_sizes;
@@ -619,41 +518,36 @@ int4 accumulate_sizes(const torch::Tensor& codebook_partition_sizes)
 }  // namespace aqlm
 }  // namespace vllm
 
-
+torch::Tensor aqlm_gemm(const torch::Tensor& input, const torch::Tensor& codes,
-torch::Tensor aqlm_gemm(
-  const torch::Tensor& input,
-  const torch::Tensor& codes,
                         const torch::Tensor& codebooks,
                         const torch::Tensor& scales,
                         const torch::Tensor& codebook_partition_sizes,
-  const std::optional<torch::Tensor>& bias
+                        const std::optional<torch::Tensor>& bias) {
-)
+  int4 cumulative_sizes =
-{
+      vllm::aqlm::accumulate_sizes(codebook_partition_sizes);
-  int4 cumulative_sizes = vllm::aqlm::accumulate_sizes(codebook_partition_sizes);
 
   int const nbooks = codebooks.size(0) / codebook_partition_sizes.size(0);
   int const entries = codebooks.size(1);
 
-  if (nbooks == 1 && entries == (1 << 16))
+  if (nbooks == 1 && entries == (1 << 16)) {
-  { 
+    return vllm::aqlm::code1x16_matmat(input, codes, codebooks, scales,
-    return vllm::aqlm::code1x16_matmat(input, codes, codebooks, scales, cumulative_sizes, bias);
+                                       cumulative_sizes, bias);
   }
-  if (nbooks == 2 && entries == (1 << 8))
+  if (nbooks == 2 && entries == (1 << 8)) {
-  {
+    return vllm::aqlm::code2x8_matmat(input, codes, codebooks, scales,
-    return vllm::aqlm::code2x8_matmat(input, codes, codebooks, scales, cumulative_sizes, bias);
+                                      cumulative_sizes, bias);
   }
 
-  TORCH_CHECK(false, "AQLM with ", nbooks, " codebooks and ", entries, " entries is not currently supported.")
+  TORCH_CHECK(false, "AQLM with ", nbooks, " codebooks and ", entries,
+              " entries is not currently supported.")
   return {};
 }
 
-torch::Tensor aqlm_dequant(
+torch::Tensor aqlm_dequant(const torch::Tensor& codes,
-  const torch::Tensor& codes,
                            const torch::Tensor& codebooks,
-  const torch::Tensor& codebook_partition_sizes
+                           const torch::Tensor& codebook_partition_sizes) {
-)
+  int4 cumulative_sizes =
-{
+      vllm::aqlm::accumulate_sizes(codebook_partition_sizes);
-  int4 cumulative_sizes = vllm::aqlm::accumulate_sizes(codebook_partition_sizes);
 
   int const nbooks = codebooks.size(0) / codebook_partition_sizes.size(0);
   int const entries = codebooks.size(1);
@@ -670,43 +564,35 @@ torch::Tensor aqlm_dequant(
   auto weights = torch::empty({out_features, in_features},
                               torch::TensorOptions()
                                   .dtype(codebooks.dtype())
-      .device(codebooks.device())
+                                  .device(codebooks.device()));
-  );
 
-  if (nbooks == 1 && entries == (1 << 16))
+  if (nbooks == 1 && entries == (1 << 16)) {
-  {
+    vllm::aqlm::code1x16_dequant_cuda(codes.data_ptr(), weights.data_ptr(),
-    vllm::aqlm::code1x16_dequant_cuda(
+                                      codebooks.data_ptr(), out_features,
-      codes.data_ptr(),
+                                      in_features, cumulative_sizes,
-      weights.data_ptr(),
-      codebooks.data_ptr(),
-      out_features,
-      in_features,
-      cumulative_sizes,
                                       vllm::aqlm::codebook_stride(codebooks));
 
-    // if you wanted to flip to scaling the weights, (though it's 30%-ish slower and not consistent with gemv implementation.)
+    // if you wanted to flip to scaling the weights, (though it's 30%-ish slower
-    // weights *= scales.index({"...", 0, 0});
+    // and not consistent with gemv implementation.) weights *=
+    // scales.index({"...", 0, 0});
 
     return weights;
   }
 
-  if (nbooks == 2 && entries == (1 << 8))
+  if (nbooks == 2 && entries == (1 << 8)) {
-  {
+    vllm::aqlm::code2x8_dequant_cuda(codes.data_ptr(), weights.data_ptr(),
-     vllm::aqlm::code2x8_dequant_cuda(
+                                     codebooks.data_ptr(), out_features,
-        codes.data_ptr(), 
+                                     in_features, cumulative_sizes,
-        weights.data_ptr(), 
-        codebooks.data_ptr(), 
-        out_features,
-        in_features, 
-        cumulative_sizes, 
                                      vllm::aqlm::codebook_stride(codebooks));
 
-    // if you wanted to flip to scaling the weights, (though it's 30%-ish slower and not consistent with gemv implementation)
+    // if you wanted to flip to scaling the weights, (though it's 30%-ish slower
-    // weights *= scales.index({"...", 0, 0});
+    // and not consistent with gemv implementation) weights *=
+    // scales.index({"...", 0, 0});
 
     return weights;
   }
 
-  TORCH_CHECK(false, "AQLM with ", nbooks, " codebooks and ", entries, " entries is not currently supported.")
+  TORCH_CHECK(false, "AQLM with ", nbooks, " codebooks and ", entries,
+              " entries is not currently supported.")
   return {};
 }

csrc/quantization/awq/dequantize.cuh

@@ -1,11 +1,11 @@
 /*
 Adapted from https://github.com/mit-han-lab/llm-awq
-Modified from NVIDIA FasterTransformer: https://github.com/NVIDIA/FasterTransformer/blob/main/src/fastertransformer/cutlass_extensions/include/cutlass_extensions/interleaved_numeric_conversion.h
+Modified from NVIDIA FasterTransformer:
+https://github.com/NVIDIA/FasterTransformer/blob/main/src/fastertransformer/cutlass_extensions/include/cutlass_extensions/interleaved_numeric_conversion.h
 @article{lin2023awq,
-  title={AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration},
+  title={AWQ: Activation-aware Weight Quantization for LLM Compression and
-  author={Lin, Ji and Tang, Jiaming and Tang, Haotian and Yang, Shang and Dang, Xingyu and Han, Song},
+Acceleration}, author={Lin, Ji and Tang, Jiaming and Tang, Haotian and Yang,
-  journal={arXiv},
+Shang and Dang, Xingyu and Han, Song}, journal={arXiv}, year={2023}
-  year={2023}
 }
 */
 
@@ -14,8 +14,7 @@ Modified from NVIDIA FasterTransformer: https://github.com/NVIDIA/FasterTransfor
 namespace vllm {
 namespace awq {
 
-__device__ uint4 dequantize_s4_to_fp16x2(uint32_t const& source)
+__device__ uint4 dequantize_s4_to_fp16x2(uint32_t const& source) {
-{
 #if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 750
   assert(false);
 #else
@@ -30,33 +29,40 @@ __device__ uint4 dequantize_s4_to_fp16x2(uint32_t const& source)
   static constexpr uint32_t TOP_MASK = 0x00f000f0;
   static constexpr uint32_t I4s_TO_F16s_MAGIC_NUM = 0x64006400;
 
-    // Note that the entire sequence only requires 1 shift instruction. This is thanks to the register packing
+  // Note that the entire sequence only requires 1 shift instruction. This is
-    // format and the fact that we force our integers to be unsigned, and account for this in the fp16 subtractions.
+  // thanks to the register packing format and the fact that we force our
-    // In addition, I exploit the fact that sub and fma have the same throughput in order to convert elt_23 and
+  // integers to be unsigned, and account for this in the fp16 subtractions. In
-    // elt_67 to fp16 without having to shift them to the bottom bits before hand.
+  // addition, I exploit the fact that sub and fma have the same throughput in
+  // order to convert elt_23 and elt_67 to fp16 without having to shift them to
+  // the bottom bits before hand.
 
-    // Shift right by 8 to now consider elt_45 and elt_67. Issue first to hide RAW dependency if we issue
+  // Shift right by 8 to now consider elt_45 and elt_67. Issue first to hide RAW
-    // immediately before required.
+  // dependency if we issue immediately before required.
   const uint32_t top_i4s = i4s >> 8;
   // Extract elt_01 - (i4s & 0x000f000f) | 0x64006400
   asm volatile("lop3.b32 %0, %1, %2, %3, %4;\n"
                : "=r"(h[0])
-                    : "r"(i4s), "n"(BOTTOM_MASK), "n"(I4s_TO_F16s_MAGIC_NUM), "n"(immLut));
+               : "r"(i4s), "n"(BOTTOM_MASK), "n"(I4s_TO_F16s_MAGIC_NUM),
+                 "n"(immLut));
   // Extract elt_23 (i4s & 0x00f000f0) | 0x64006400
   asm volatile("lop3.b32 %0, %1, %2, %3, %4;\n"
                : "=r"(h[1])
-                    : "r"(i4s), "n"(TOP_MASK), "n"(I4s_TO_F16s_MAGIC_NUM), "n"(immLut));
+               : "r"(i4s), "n"(TOP_MASK), "n"(I4s_TO_F16s_MAGIC_NUM),
+                 "n"(immLut));
   // Extract elt_45 (top_i4s & 0x000f000f) | 0x64006400
   asm volatile("lop3.b32 %0, %1, %2, %3, %4;\n"
                : "=r"(h[2])
-                    : "r"(top_i4s), "n"(BOTTOM_MASK), "n"(I4s_TO_F16s_MAGIC_NUM), "n"(immLut));
+               : "r"(top_i4s), "n"(BOTTOM_MASK), "n"(I4s_TO_F16s_MAGIC_NUM),
+                 "n"(immLut));
   // Extract elt_67 (top_i4s & 0x00f000f0) | 0x64006400
   asm volatile("lop3.b32 %0, %1, %2, %3, %4;\n"
                : "=r"(h[3])
-                    : "r"(top_i4s), "n"(TOP_MASK), "n"(I4s_TO_F16s_MAGIC_NUM), "n"(immLut));
+               : "r"(top_i4s), "n"(TOP_MASK), "n"(I4s_TO_F16s_MAGIC_NUM),
+                 "n"(immLut));
 
-    // I use inline PTX below because I am not sure if the compiler will emit float2half instructions if I use the
+  // I use inline PTX below because I am not sure if the compiler will emit
-    // half2 ctor. In this case, I chose performance reliability over code readability.
+  // float2half instructions if I use the half2 ctor. In this case, I chose
+  // performance reliability over code readability.
 
   // This is the half2 {1032, 1032} represented as an integer.
   // static constexpr uint32_t FP16_TOP_MAGIC_NUM = 0x64086408;
@@ -71,13 +77,21 @@ __device__ uint4 dequantize_s4_to_fp16x2(uint32_t const& source)
 
   // Finally, we construct the output numbers.
   // Convert elt_01
-    asm volatile("sub.f16x2 %0, %1, %2;\n" : "=r"(h[0]) : "r"(h[0]), "r"(FP16_TOP_MAGIC_NUM));
+  asm volatile("sub.f16x2 %0, %1, %2;\n"
+               : "=r"(h[0])
+               : "r"(h[0]), "r"(FP16_TOP_MAGIC_NUM));
   // Convert elt_23
-    asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n" : "=r"(h[1]) : "r"(h[1]), "r"(ONE_SIXTEENTH), "r"(NEG_64));
+  asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n"
+               : "=r"(h[1])
+               : "r"(h[1]), "r"(ONE_SIXTEENTH), "r"(NEG_64));
   // Convert elt_45
-    asm volatile("sub.f16x2 %0, %1, %2;\n" : "=r"(h[2]) : "r"(h[2]), "r"(FP16_TOP_MAGIC_NUM));
+  asm volatile("sub.f16x2 %0, %1, %2;\n"
+               : "=r"(h[2])
+               : "r"(h[2]), "r"(FP16_TOP_MAGIC_NUM));
   // Convert elt_67
-    asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n" : "=r"(h[3]) : "r"(h[3]), "r"(ONE_SIXTEENTH), "r"(NEG_64));
+  asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n"
+               : "=r"(h[3])
+               : "r"(h[3]), "r"(ONE_SIXTEENTH), "r"(NEG_64));
 
   return result;
 #endif

csrc/quantization/awq/gemm_kernels.cu

@@ -1,15 +1,13 @@
 /*
 Adapted from https://github.com/mit-han-lab/llm-awq
 @article{lin2023awq,
-  title={AWQ: Activation-aware Weight Quantization for LLM Compression and Acceleration},
+  title={AWQ: Activation-aware Weight Quantization for LLM Compression and
-  author={Lin, Ji and Tang, Jiaming and Tang, Haotian and Yang, Shang and Dang, Xingyu and Han, Song},
+Acceleration}, author={Lin, Ji and Tang, Jiaming and Tang, Haotian and Yang,
-  journal={arXiv},
+Shang and Dang, Xingyu and Han, Song}, journal={arXiv}, year={2023}
-  year={2023}
 }
  */
 
-
+#include <torch/all.h>
-#include <torch/extension.h>
 #include <c10/cuda/CUDAGuard.h>
 
 #include "dequantize.cuh"
@@ -20,26 +18,20 @@ namespace vllm {
 namespace awq {
 
 // Pack two half values.
-static inline __device__ __host__ unsigned
+static inline __device__ __host__ unsigned __pack_half2(const half x,
-__pack_half2(const half x, const half y) {
+                                                        const half y) {
   unsigned v0 = *((unsigned short*)&x);
   unsigned v1 = *((unsigned short*)&y);
   return (v1 << 16) | v0;
 }
 
 template <int N>
-__global__ void __launch_bounds__(64) gemm_forward_4bit_cuda_m16nXk32(
+__global__ void __launch_bounds__(64)
-  int G,
+    gemm_forward_4bit_cuda_m16nXk32(int G, int split_k_iters,
-  int split_k_iters,
+                                    half* __restrict__ A, int* __restrict__ B,
-  half* __restrict__ A,
-  int* __restrict__ B,
                                     half* __restrict__ scaling_factors,
-  int* __restrict__ zeros,
+                                    int* __restrict__ zeros, int M, int IC,
-  int M,
+                                    int OC, half* __restrict__ C) {
-  int IC,
-  int OC,
-  half* __restrict__ C)
-{
   // Only support matrix n = 64 or 128
   assert(N == 64 || N == 128);
 #if defined(__CUDA_ARCH__) && __CUDA_ARCH__ < 750
@@ -70,43 +62,46 @@ __global__ void __launch_bounds__(64) gemm_forward_4bit_cuda_m16nXk32(
   static constexpr int row_stride = 2 * 32 * 8 / N;
   bool ld_zero_flag = (threadIdx.y * 32 + threadIdx.x) * 8 < N;
   // TODO: Haotian: blockIdx_y / j_factors1 in A loading to support bsz > 16
-  bool ld_A_flag = (blockIdx_y / j_factors1 * 16 + threadIdx.y * row_stride_warp + threadIdx.x * 8 / 32) < M;     // threadIdx.y is warp_id
+  bool ld_A_flag =
+      (blockIdx_y / j_factors1 * 16 + threadIdx.y * row_stride_warp +
+       threadIdx.x * 8 / 32) < M;  // threadIdx.y is warp_id
   // bool wb_C_flag = (threadIdx.x / 4) < M;
 
-  half* A_ptr = A
+  half* A_ptr =
-                + (((int)blockIdx_y) / j_factors1 * 16 + (((int)threadIdx.y) * row_stride_warp) + ((int)threadIdx.x) / (32 / 8)) * IC
+      A +
-                + (((int)threadIdx.x) % (32 / 8)) * 8;
+      (((int)blockIdx_y) / j_factors1 * 16 +
+       (((int)threadIdx.y) * row_stride_warp) + ((int)threadIdx.x) / (32 / 8)) *
+          IC +
+      (((int)threadIdx.x) % (32 / 8)) * 8;
 
-  int* B_ptr = B
+  int* B_ptr = B + ((int)threadIdx.y) * (OC / 8) * (256 / N) +
-            + ((int)threadIdx.y) * (OC / 8) * (256 / N)
+               (((int)threadIdx.x) / (N / 8)) * (OC / 8) +
-            + (((int)threadIdx.x) / (N / 8)) * (OC / 8)
+               (((int)blockIdx_y) % j_factors1) * (N / 8) +
-            + (((int)blockIdx_y) % j_factors1) * (N / 8)
+               (((int)threadIdx.x) % (N / 8)) * 1;
-            + (((int)threadIdx.x) % (N / 8)) * 1;
   // Why * 1 in the above line?
 
-  half* A_shared_ptr = A_shared
+  half* A_shared_ptr = A_shared +
-                    + ((int)threadIdx.y) * row_stride_warp * (32 + 8)
+                       ((int)threadIdx.y) * row_stride_warp * (32 + 8) +
-                    + (((int)threadIdx.x) / (32 / 8)) * (32 + 8)
+                       (((int)threadIdx.x) / (32 / 8)) * (32 + 8) +
-                    + (((int)threadIdx.x) % (32 / 8) ) * 8;
+                       (((int)threadIdx.x) % (32 / 8)) * 8;
 
-  half* B_shared_ptr = B_shared
+  half* B_shared_ptr = B_shared +
-                    + ((int)threadIdx.y) * (row_stride / 2) * (N + 8)
+                       ((int)threadIdx.y) * (row_stride / 2) * (N + 8) +
-                    + (((int)threadIdx.x) / (N / 8)) * (N + 8)
+                       (((int)threadIdx.x) / (N / 8)) * (N + 8) +
-                    + (((int)threadIdx.x) % (N / 8)) * 8;
+                       (((int)threadIdx.x) % (N / 8)) * 8;
 
-  int* zeros_ptr = zeros
+  int* zeros_ptr = zeros + (((int)blockIdx_y) % j_factors1) * (N / 8) +
-                + (((int)blockIdx_y) % j_factors1) * (N / 8)
+                   ((int)threadIdx.x) % (N / 8);
-                + ((int)threadIdx.x) % (N / 8);
 
-  half* scaling_factors_ptr = scaling_factors
+  half* scaling_factors_ptr = scaling_factors +
-                            + (((int)blockIdx_y) % j_factors1) * N
+                              (((int)blockIdx_y) % j_factors1) * N +
-                            + (((int)threadIdx.x) % (N / 8)) * 8;
+                              (((int)threadIdx.x) % (N / 8)) * 8;
 
-  half* C_ptr = C
+  half* C_ptr =
-              + static_cast<long long>(blockIdx_z) * M * OC        // blockIdz.x -> split_k dim
+      C +
-              + (((int)blockIdx_y) % j_factors1) * N
+      static_cast<long long>(blockIdx_z) * M * OC  // blockIdz.x -> split_k dim
-              + ((int)threadIdx.y) * (N / 2)
+      + (((int)blockIdx_y) % j_factors1) * N + ((int)threadIdx.y) * (N / 2) +
-              + (((int)threadIdx.x) % 4) * 2;
+      (((int)threadIdx.x) % 4) * 2;
 
   // preload s.f. and zeros
   int k_bound = (IC / 32 + split_k_iters - 1) / split_k_iters;
@@ -115,57 +110,83 @@ __global__ void __launch_bounds__(64) gemm_forward_4bit_cuda_m16nXk32(
     int k_0_0 = _k_0_0 * split_k_iters + blockIdx_z;
     __syncthreads();
     // TODO: Haotian: blockIdx_y / j_factors1 in A loading to support bsz > 16
-    if (ld_A_flag)
+    if (ld_A_flag) {
-    {
       *(uint4*)(A_shared_ptr) = *(uint4*)(A_ptr + (k_0_0 * 32));
-    }
+    } else {
-    else
-    {
       *(uint4*)(A_shared_ptr) = make_uint4(0, 0, 0, 0);
     }
 
     // for (int ax0_ax1_fused_0 = 0; ax0_ax1_fused_0 < 2; ++ax0_ax1_fused_0) {
     uint32_t zeros_loaded = *(uint32_t*)(zeros_ptr + k_0_0 * 32 / G * (OC / 8));
     uint4 B_loaded_zero = dequantize_s4_to_fp16x2(zeros_loaded);
-    uint4 B_loaded_scale = *(uint4*)(scaling_factors_ptr + k_0_0 * 32 / G * (OC));
+    uint4 B_loaded_scale =
+        *(uint4*)(scaling_factors_ptr + k_0_0 * 32 / G * (OC));
     /*
-    if (blockIdx_z == 0 && blockIdx_y == 0 && k_0_0 == 0 && threadIdx.x == 0 && threadIdx.y == 0){
+    if (blockIdx_z == 0 && blockIdx_y == 0 && k_0_0 == 0 && threadIdx.x == 0 &&
-      printf("%x %x %x %x %x %x %x %x\n", B_loaded_scale.x, B_loaded_scale.y, B_loaded_scale.z, B_loaded_scale.w, B_loaded_zero.x, B_loaded_zero.y, B_loaded_zero.z, B_loaded_zero.w);
+    threadIdx.y == 0){ printf("%x %x %x %x %x %x %x %x\n", B_loaded_scale.x,
+    B_loaded_scale.y, B_loaded_scale.z, B_loaded_scale.w, B_loaded_zero.x,
+    B_loaded_zero.y, B_loaded_zero.z, B_loaded_zero.w);
     }
     */
     // uint4 B_loaded_scale = make_uint4(0, 0, 0, 0);
     int* B_ptr_local = B_ptr + k_0_0 * 32 * (OC / 8);
 
     for (int ax0_ax1_fused_0 = 0; ax0_ax1_fused_0 < N / 16; ++ax0_ax1_fused_0) {
-
       // B: 32 x 136 (128+8) float16
       // each warp: 32 x 4
-      // each thr: read 32 bit -> convert to 8xFP16 (a UINT4) -> scale and minus zero -> WB UINT4
+      // each thr: read 32 bit -> convert to 8xFP16 (a UINT4) -> scale and minus
-      // *(uint4*)(B_shared + ((((ax0_ax1_fused_0 * 544) + (((int)threadIdx.y) * 272)) + ((((int)threadIdx.x) >> 4) * 136)) + ((((int)threadIdx.x) & 15) * 8))) = *(uint4*)(B + ((((((k_0_0 * 163840) + (ax0_ax1_fused_0 * 20480)) + (((int)threadIdx.y) * 10240)) + ((((int)threadIdx.x) >> 4) * 5120)) + (((int)blockIdx_y) * 128)) + ((((int)threadIdx.x) & 15) * 8)));
+      // zero -> WB UINT4
-      // row stride in shared memory: (NWARPS * 32 * 8 / cta_N)
+      // *(uint4*)(B_shared + ((((ax0_ax1_fused_0 * 544) + (((int)threadIdx.y) *
-      uint32_t B_loaded = *(uint32_t*)(B_ptr_local + ax0_ax1_fused_0 * row_stride * (OC / 8));
+      // 272)) + ((((int)threadIdx.x) >> 4) * 136)) + ((((int)threadIdx.x) & 15)
+      // * 8))) = *(uint4*)(B + ((((((k_0_0 * 163840) + (ax0_ax1_fused_0 *
+      // 20480)) + (((int)threadIdx.y) * 10240)) + ((((int)threadIdx.x) >> 4) *
+      // 5120)) + (((int)blockIdx_y) * 128)) + ((((int)threadIdx.x) & 15) *
+      // 8))); row stride in shared memory: (NWARPS * 32 * 8 / cta_N)
+      uint32_t B_loaded =
+          *(uint32_t*)(B_ptr_local + ax0_ax1_fused_0 * row_stride * (OC / 8));
       uint4 B_loaded_fp16 = dequantize_s4_to_fp16x2(B_loaded);
-      //uint4 B_loaded_zero = *(uint4*)(zeros_shared + (threadIdx.x % (cta_N / 8)) * 8);
+      // uint4 B_loaded_zero = *(uint4*)(zeros_shared + (threadIdx.x % (cta_N /
+      // 8)) * 8);
 
-      // uint4 B_loaded_scale = *(uint4*)(scaling_factors_shared + (threadIdx.x % (cta_N / 8)) * 8);
+      // uint4 B_loaded_scale = *(uint4*)(scaling_factors_shared + (threadIdx.x
+      // % (cta_N / 8)) * 8);
       // - zero and * scale
-      // TODO (Haotian): can save 4 assembly instructions if sormulate as deq = q * scale - zero * scale.
+      // TODO (Haotian): can save 4 assembly instructions if sormulate as deq =
-      asm volatile("sub.f16x2 %0, %1, %2;\n" : "=r"(B_loaded_fp16.x) : "r"(B_loaded_fp16.x), "r"(B_loaded_zero.x));
+      // q * scale - zero * scale.
-      asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n" : "=r"(B_loaded_fp16.x) : "r"(B_loaded_fp16.x), "r"(B_loaded_scale.x), "r"(ZERO));
+      asm volatile("sub.f16x2 %0, %1, %2;\n"
-      asm volatile("sub.f16x2 %0, %1, %2;\n" : "=r"(B_loaded_fp16.y) : "r"(B_loaded_fp16.y), "r"(B_loaded_zero.y));
+                   : "=r"(B_loaded_fp16.x)
-      asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n" : "=r"(B_loaded_fp16.y) : "r"(B_loaded_fp16.y), "r"(B_loaded_scale.y), "r"(ZERO));
+                   : "r"(B_loaded_fp16.x), "r"(B_loaded_zero.x));
-      asm volatile("sub.f16x2 %0, %1, %2;\n" : "=r"(B_loaded_fp16.z) : "r"(B_loaded_fp16.z), "r"(B_loaded_zero.z));
+      asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n"
-      asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n" : "=r"(B_loaded_fp16.z) : "r"(B_loaded_fp16.z), "r"(B_loaded_scale.z), "r"(ZERO));
+                   : "=r"(B_loaded_fp16.x)
-      asm volatile("sub.f16x2 %0, %1, %2;\n" : "=r"(B_loaded_fp16.w) : "r"(B_loaded_fp16.w), "r"(B_loaded_zero.w));
+                   : "r"(B_loaded_fp16.x), "r"(B_loaded_scale.x), "r"(ZERO));
-      asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n" : "=r"(B_loaded_fp16.w) : "r"(B_loaded_fp16.w), "r"(B_loaded_scale.w), "r"(ZERO));
+      asm volatile("sub.f16x2 %0, %1, %2;\n"
+                   : "=r"(B_loaded_fp16.y)
+                   : "r"(B_loaded_fp16.y), "r"(B_loaded_zero.y));
+      asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n"
+                   : "=r"(B_loaded_fp16.y)
+                   : "r"(B_loaded_fp16.y), "r"(B_loaded_scale.y), "r"(ZERO));
+      asm volatile("sub.f16x2 %0, %1, %2;\n"
+                   : "=r"(B_loaded_fp16.z)
+                   : "r"(B_loaded_fp16.z), "r"(B_loaded_zero.z));
+      asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n"
+                   : "=r"(B_loaded_fp16.z)
+                   : "r"(B_loaded_fp16.z), "r"(B_loaded_scale.z), "r"(ZERO));
+      asm volatile("sub.f16x2 %0, %1, %2;\n"
+                   : "=r"(B_loaded_fp16.w)
+                   : "r"(B_loaded_fp16.w), "r"(B_loaded_zero.w));
+      asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n"
+                   : "=r"(B_loaded_fp16.w)
+                   : "r"(B_loaded_fp16.w), "r"(B_loaded_scale.w), "r"(ZERO));
       /*
-      if (ax0_ax1_fused_0 == 0 && blockIdx_z == 0 && blockIdx_y == 0 && k_0_0 == 0 && threadIdx.x == 17 && threadIdx.y == 0){
+      if (ax0_ax1_fused_0 == 0 && blockIdx_z == 0 && blockIdx_y == 0 && k_0_0 ==
-        printf("[x] %X %X %X %X\n", B_loaded_fp16.x, B_loaded_fp16.y, B_loaded_fp16.z, B_loaded_fp16.w);
+      0 && threadIdx.x == 17 && threadIdx.y == 0){ printf("[x] %X %X %X %X\n",
+      B_loaded_fp16.x, B_loaded_fp16.y, B_loaded_fp16.z, B_loaded_fp16.w);
       }
       */
 
       // write back
-      *(uint4*)(B_shared_ptr + ax0_ax1_fused_0 * row_stride * (N + 8)) = B_loaded_fp16;
+      *(uint4*)(B_shared_ptr + ax0_ax1_fused_0 * row_stride * (N + 8)) =
+          B_loaded_fp16;
     }
     __syncthreads();
 
@@ -173,34 +194,43 @@ __global__ void __launch_bounds__(64) gemm_forward_4bit_cuda_m16nXk32(
       {
         unsigned int addr;
         __asm__ __volatile__(
-          "{ .reg .u64 addr; cvta.to.shared.u64 addr, %1; cvt.u32.u64 %0, addr; }\n"
+            "{ .reg .u64 addr; cvta.to.shared.u64 addr, %1; cvt.u32.u64 %0, "
+            "addr; }\n"
             : "=r"(addr)
-          : "l"((void *)((&(A_shared[(k_0_1 * 16)])) + (((((int)threadIdx.x) & 15) * 40) + ((((int)threadIdx.x) >> 4) * 8))))
+            : "l"((void*)((&(A_shared[(k_0_1 * 16)])) +
-        );
+                          (((((int)threadIdx.x) & 15) * 40) +
-
+                           ((((int)threadIdx.x) >> 4) * 8)))));
 
         __asm__ __volatile__(
             "ldmatrix.sync.aligned.m8n8.x4.shared.b16"
             "{%0, %1, %2, %3}, [%4];\n"
-          : "=r"(((unsigned *)(A_shared_warp + 0))[0]), "=r"(((unsigned *)(A_shared_warp + 0))[1]), "=r"(((unsigned *)(A_shared_warp + 0))[2]), "=r"(((unsigned *)(A_shared_warp + 0))[3])
+            : "=r"(((unsigned*)(A_shared_warp + 0))[0]),
-          : "r"(addr)
+              "=r"(((unsigned*)(A_shared_warp + 0))[1]),
-        );
+              "=r"(((unsigned*)(A_shared_warp + 0))[2]),
+              "=r"(((unsigned*)(A_shared_warp + 0))[3])
+            : "r"(addr));
       }
 
       for (int ax1_0 = 0; ax1_0 < N / 32; ++ax1_0) {
         {
           unsigned int addr;
           __asm__ __volatile__(
-            "{ .reg .u64 addr; cvta.to.shared.u64 addr, %1; cvt.u32.u64 %0, addr; }\n"
+              "{ .reg .u64 addr; cvta.to.shared.u64 addr, %1; cvt.u32.u64 %0, "
+              "addr; }\n"
               : "=r"(addr)
-            : "l"((void *)((&(B_shared[(((k_0_1 * (N * 16 + 128)) + (((int)threadIdx.y) * (N / 2))) + (ax1_0 * 16))])) + (((((int)threadIdx.x) & 15) * (N + 8)) + ((((int)threadIdx.x) >> 4) * 8))))
+              : "l"((void*)((&(B_shared[(((k_0_1 * (N * 16 + 128)) +
-          );
+                                          (((int)threadIdx.y) * (N / 2))) +
+                                         (ax1_0 * 16))])) +
+                            (((((int)threadIdx.x) & 15) * (N + 8)) +
+                             ((((int)threadIdx.x) >> 4) * 8)))));
           __asm__ __volatile__(
               "ldmatrix.sync.aligned.m8n8.x4.trans.shared.b16"
               "{%0, %1, %2, %3}, [%4];\n"
-            : "=r"(((unsigned *)(B_shared_warp + (ax1_0 * 8)))[0]), "=r"(((unsigned *)(B_shared_warp + (ax1_0 * 8)))[1]), "=r"(((unsigned *)(B_shared_warp + (ax1_0 * 8)))[2]), "=r"(((unsigned *)(B_shared_warp + (ax1_0 * 8)))[3])
+              : "=r"(((unsigned*)(B_shared_warp + (ax1_0 * 8)))[0]),
-            : "r"(addr)
+                "=r"(((unsigned*)(B_shared_warp + (ax1_0 * 8)))[1]),
-          );
+                "=r"(((unsigned*)(B_shared_warp + (ax1_0 * 8)))[2]),
+                "=r"(((unsigned*)(B_shared_warp + (ax1_0 * 8)))[3])
+              : "r"(addr));
         }
       }
       for (int j_0_4 = 0; j_0_4 < N / 32; ++j_0_4) {
@@ -209,48 +239,110 @@ __global__ void __launch_bounds__(64) gemm_forward_4bit_cuda_m16nXk32(
           __asm__ __volatile__(
               "mma.sync.aligned.m16n8k8.row.col.f32.f16.f16.f32"
               "{%0, %1, %2, %3}, {%4, %5}, {%6}, {%7, %8, %9, %10};\n"
-            :  "=f"(((float *)(C_warp + (j_0_4 * 8)))[0]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[1]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[2]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[3])
+              : "=f"(((float*)(C_warp + (j_0_4 * 8)))[0]),
-            : "r"(((unsigned *)(A_shared_warp + 0))[0]), "r"(((unsigned *)(A_shared_warp + 0))[1]), "r"(((unsigned *)(B_shared_warp + (j_0_4 * 8)))[0]), "f"(((float *)(C_warp + (j_0_4 * 8)))[0]), "f"(((float *)(C_warp + (j_0_4 * 8)))[1]), "f"(((float *)(C_warp + (j_0_4 * 8)))[2]), "f"(((float *)(C_warp + (j_0_4 * 8)))[3]));
+                "=f"(((float*)(C_warp + (j_0_4 * 8)))[1]),
+                "=f"(((float*)(C_warp + (j_0_4 * 8)))[2]),
+                "=f"(((float*)(C_warp + (j_0_4 * 8)))[3])
+              : "r"(((unsigned*)(A_shared_warp + 0))[0]),
+                "r"(((unsigned*)(A_shared_warp + 0))[1]),
+                "r"(((unsigned*)(B_shared_warp + (j_0_4 * 8)))[0]),
+                "f"(((float*)(C_warp + (j_0_4 * 8)))[0]),
+                "f"(((float*)(C_warp + (j_0_4 * 8)))[1]),
+                "f"(((float*)(C_warp + (j_0_4 * 8)))[2]),
+                "f"(((float*)(C_warp + (j_0_4 * 8)))[3]));
         }
 
         {
           __asm__ __volatile__(
               "mma.sync.aligned.m16n8k8.row.col.f32.f16.f16.f32"
               "{%0, %1, %2, %3}, {%4, %5}, {%6}, {%7, %8, %9, %10};\n"
-            :  "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3])
+              : "=f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[0]),
-            : "r"(((unsigned *)(A_shared_warp + 0))[0]), "r"(((unsigned *)(A_shared_warp + 0))[1]), "r"(((unsigned *)(B_shared_warp + ((j_0_4 * 8) + 4)))[0]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3]));
+                "=f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[1]),
+                "=f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[2]),
+                "=f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[3])
+              : "r"(((unsigned*)(A_shared_warp + 0))[0]),
+                "r"(((unsigned*)(A_shared_warp + 0))[1]),
+                "r"(((unsigned*)(B_shared_warp + ((j_0_4 * 8) + 4)))[0]),
+                "f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[0]),
+                "f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[1]),
+                "f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[2]),
+                "f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[3]));
         }
 
         {
           __asm__ __volatile__(
               "mma.sync.aligned.m16n8k8.row.col.f32.f16.f16.f32"
               "{%0, %1, %2, %3}, {%4, %5}, {%6}, {%7, %8, %9, %10};\n"
-            :  "=f"(((float *)(C_warp + (j_0_4 * 8)))[0]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[1]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[2]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[3])
+              : "=f"(((float*)(C_warp + (j_0_4 * 8)))[0]),
-            : "r"(((unsigned *)(A_shared_warp + 0))[2]), "r"(((unsigned *)(A_shared_warp + 0))[3]), "r"(((unsigned *)(B_shared_warp + (j_0_4 * 8)))[1]), "f"(((float *)(C_warp + (j_0_4 * 8)))[0]), "f"(((float *)(C_warp + (j_0_4 * 8)))[1]), "f"(((float *)(C_warp + (j_0_4 * 8)))[2]), "f"(((float *)(C_warp + (j_0_4 * 8)))[3]));
+                "=f"(((float*)(C_warp + (j_0_4 * 8)))[1]),
+                "=f"(((float*)(C_warp + (j_0_4 * 8)))[2]),
+                "=f"(((float*)(C_warp + (j_0_4 * 8)))[3])
+              : "r"(((unsigned*)(A_shared_warp + 0))[2]),
+                "r"(((unsigned*)(A_shared_warp + 0))[3]),
+                "r"(((unsigned*)(B_shared_warp + (j_0_4 * 8)))[1]),
+                "f"(((float*)(C_warp + (j_0_4 * 8)))[0]),
+                "f"(((float*)(C_warp + (j_0_4 * 8)))[1]),
+                "f"(((float*)(C_warp + (j_0_4 * 8)))[2]),
+                "f"(((float*)(C_warp + (j_0_4 * 8)))[3]));
         }
 
         {
           __asm__ __volatile__(
               "mma.sync.aligned.m16n8k8.row.col.f32.f16.f16.f32"
               "{%0, %1, %2, %3}, {%4, %5}, {%6}, {%7, %8, %9, %10};\n"
-            :  "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3])
+              : "=f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[0]),
-            : "r"(((unsigned *)(A_shared_warp + 0))[2]), "r"(((unsigned *)(A_shared_warp + 0))[3]), "r"(((unsigned *)(B_shared_warp + ((j_0_4 * 8) + 4)))[1]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3]));
+                "=f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[1]),
+                "=f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[2]),
+                "=f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[3])
+              : "r"(((unsigned*)(A_shared_warp + 0))[2]),
+                "r"(((unsigned*)(A_shared_warp + 0))[3]),
+                "r"(((unsigned*)(B_shared_warp + ((j_0_4 * 8) + 4)))[1]),
+                "f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[0]),
+                "f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[1]),
+                "f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[2]),
+                "f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[3]));
         }
   #else
         {
           __asm__ __volatile__(
               "mma.sync.aligned.m16n8k16.row.col.f32.f16.f16.f32"
-            "{%0, %1, %2, %3}, {%4, %5, %6, %7}, {%8, %9}, {%10, %11, %12, %13};\n"
+              "{%0, %1, %2, %3}, {%4, %5, %6, %7}, {%8, %9}, {%10, %11, %12, "
-            :  "=f"(((float *)(C_warp + (j_0_4 * 8)))[0]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[1]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[2]), "=f"(((float *)(C_warp + (j_0_4 * 8)))[3])
+              "%13};\n"
-            : "r"(((unsigned *)(A_shared_warp + 0))[0]), "r"(((unsigned *)(A_shared_warp + 0))[1]), "r"(((unsigned *)(A_shared_warp + 0))[2]), "r"(((unsigned *)(A_shared_warp + 0))[3]), "r"(((unsigned *)(B_shared_warp + (j_0_4 * 8)))[0]), "r"(((unsigned *)(B_shared_warp + (j_0_4 * 8)))[1]), "f"(((float *)(C_warp + (j_0_4 * 8)))[0]), "f"(((float *)(C_warp + (j_0_4 * 8)))[1]), "f"(((float *)(C_warp + (j_0_4 * 8)))[2]), "f"(((float *)(C_warp + (j_0_4 * 8)))[3]));
+              : "=f"(((float*)(C_warp + (j_0_4 * 8)))[0]),
+                "=f"(((float*)(C_warp + (j_0_4 * 8)))[1]),
+                "=f"(((float*)(C_warp + (j_0_4 * 8)))[2]),
+                "=f"(((float*)(C_warp + (j_0_4 * 8)))[3])
+              : "r"(((unsigned*)(A_shared_warp + 0))[0]),
+                "r"(((unsigned*)(A_shared_warp + 0))[1]),
+                "r"(((unsigned*)(A_shared_warp + 0))[2]),
+                "r"(((unsigned*)(A_shared_warp + 0))[3]),
+                "r"(((unsigned*)(B_shared_warp + (j_0_4 * 8)))[0]),
+                "r"(((unsigned*)(B_shared_warp + (j_0_4 * 8)))[1]),
+                "f"(((float*)(C_warp + (j_0_4 * 8)))[0]),
+                "f"(((float*)(C_warp + (j_0_4 * 8)))[1]),
+                "f"(((float*)(C_warp + (j_0_4 * 8)))[2]),
+                "f"(((float*)(C_warp + (j_0_4 * 8)))[3]));
         }
 
         {
           __asm__ __volatile__(
               "mma.sync.aligned.m16n8k16.row.col.f32.f16.f16.f32"
-            "{%0, %1, %2, %3}, {%4, %5, %6, %7}, {%8, %9}, {%10, %11, %12, %13};\n"
+              "{%0, %1, %2, %3}, {%4, %5, %6, %7}, {%8, %9}, {%10, %11, %12, "
-            :  "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "=f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3])
+              "%13};\n"
-            : "r"(((unsigned *)(A_shared_warp + 0))[0]), "r"(((unsigned *)(A_shared_warp + 0))[1]), "r"(((unsigned *)(A_shared_warp + 0))[2]), "r"(((unsigned *)(A_shared_warp + 0))[3]), "r"(((unsigned *)(B_shared_warp + ((j_0_4 * 8) + 4)))[0]), "r"(((unsigned *)(B_shared_warp + ((j_0_4 * 8) + 4)))[1]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[0]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[1]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[2]), "f"(((float *)(C_warp + ((j_0_4 * 8) + 4)))[3]));
+              : "=f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[0]),
+                "=f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[1]),
+                "=f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[2]),
+                "=f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[3])
+              : "r"(((unsigned*)(A_shared_warp + 0))[0]),
+                "r"(((unsigned*)(A_shared_warp + 0))[1]),
+                "r"(((unsigned*)(A_shared_warp + 0))[2]),
+                "r"(((unsigned*)(A_shared_warp + 0))[3]),
+                "r"(((unsigned*)(B_shared_warp + ((j_0_4 * 8) + 4)))[0]),
+                "r"(((unsigned*)(B_shared_warp + ((j_0_4 * 8) + 4)))[1]),
+                "f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[0]),
+                "f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[1]),
+                "f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[2]),
+                "f"(((float*)(C_warp + ((j_0_4 * 8) + 4)))[3]));
         }
 
   #endif
@@ -261,24 +353,20 @@ __global__ void __launch_bounds__(64) gemm_forward_4bit_cuda_m16nXk32(
   // TODO: Shang: Hoist loop invariance.
   for (int ax1_0_1 = 0; ax1_0_1 < 4; ++ax1_0_1) {
     for (int local_id = 0; local_id < 8; ++local_id) {
-      int row_offset = (((int)blockIdx_y) / j_factors1) * 16 + ((int)threadIdx.x) / 4 + (local_id % 4) / 2 * 8;
+      int row_offset = (((int)blockIdx_y) / j_factors1) * 16 +
-      if (row_offset < M)
+                       ((int)threadIdx.x) / 4 + (local_id % 4) / 2 * 8;
-      {
+      if (row_offset < M) {
-        *(C_ptr + ax1_0_1 * 16 + row_offset * OC + (local_id / 4) * 8 + local_id % 2) = __float2half(C_warp[(ax1_0_1 * 8) + local_id]);
+        *(C_ptr + ax1_0_1 * 16 + row_offset * OC + (local_id / 4) * 8 +
+          local_id % 2) = __float2half(C_warp[(ax1_0_1 * 8) + local_id]);
       }
     }
   }
 #endif
 }
 
-__global__ void __launch_bounds__(64) dequantize_weights(
+__global__ void __launch_bounds__(64)
-    int* __restrict__ B,
+    dequantize_weights(int* __restrict__ B, half* __restrict__ scaling_factors,
-    half* __restrict__ scaling_factors,
+                       int* __restrict__ zeros, half* __restrict__ C, int G) {
-    int* __restrict__ zeros,
-    half* __restrict__ C,
-    int G
-)
-{
   int j_factors1 = 4;
   int row_stride2 = 4;
   int split_k_iters = 1;
@@ -310,14 +398,30 @@ __global__ void __launch_bounds__(64) dequantize_weights(
 
   uint32_t B_loaded = *(uint32_t*)B_ptr2;
   uint4 B_loaded_fp16 = dequantize_s4_to_fp16x2(B_loaded);
-  asm volatile("sub.f16x2 %0, %1, %2;\n" : "=r"(B_loaded_fp16.x) : "r"(B_loaded_fp16.x), "r"(B_loaded_zero.x));
+  asm volatile("sub.f16x2 %0, %1, %2;\n"
-  asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n" : "=r"(B_loaded_fp16.x) : "r"(B_loaded_fp16.x), "r"(B_loaded_scale.x), "r"(ZERO));
+               : "=r"(B_loaded_fp16.x)
-  asm volatile("sub.f16x2 %0, %1, %2;\n" : "=r"(B_loaded_fp16.y) : "r"(B_loaded_fp16.y), "r"(B_loaded_zero.y));
+               : "r"(B_loaded_fp16.x), "r"(B_loaded_zero.x));
-  asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n" : "=r"(B_loaded_fp16.y) : "r"(B_loaded_fp16.y), "r"(B_loaded_scale.y), "r"(ZERO));
+  asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n"
-  asm volatile("sub.f16x2 %0, %1, %2;\n" : "=r"(B_loaded_fp16.z) : "r"(B_loaded_fp16.z), "r"(B_loaded_zero.z));
+               : "=r"(B_loaded_fp16.x)
-  asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n" : "=r"(B_loaded_fp16.z) : "r"(B_loaded_fp16.z), "r"(B_loaded_scale.z), "r"(ZERO));
+               : "r"(B_loaded_fp16.x), "r"(B_loaded_scale.x), "r"(ZERO));
-  asm volatile("sub.f16x2 %0, %1, %2;\n" : "=r"(B_loaded_fp16.w) : "r"(B_loaded_fp16.w), "r"(B_loaded_zero.w));
+  asm volatile("sub.f16x2 %0, %1, %2;\n"
-  asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n" : "=r"(B_loaded_fp16.w) : "r"(B_loaded_fp16.w), "r"(B_loaded_scale.w), "r"(ZERO));
+               : "=r"(B_loaded_fp16.y)
+               : "r"(B_loaded_fp16.y), "r"(B_loaded_zero.y));
+  asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n"
+               : "=r"(B_loaded_fp16.y)
+               : "r"(B_loaded_fp16.y), "r"(B_loaded_scale.y), "r"(ZERO));
+  asm volatile("sub.f16x2 %0, %1, %2;\n"
+               : "=r"(B_loaded_fp16.z)
+               : "r"(B_loaded_fp16.z), "r"(B_loaded_zero.z));
+  asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n"
+               : "=r"(B_loaded_fp16.z)
+               : "r"(B_loaded_fp16.z), "r"(B_loaded_scale.z), "r"(ZERO));
+  asm volatile("sub.f16x2 %0, %1, %2;\n"
+               : "=r"(B_loaded_fp16.w)
+               : "r"(B_loaded_fp16.w), "r"(B_loaded_zero.w));
+  asm volatile("fma.rn.f16x2 %0, %1, %2, %3;\n"
+               : "=r"(B_loaded_fp16.w)
+               : "r"(B_loaded_fp16.w), "r"(B_loaded_scale.w), "r"(ZERO));
 
   *(uint4*)B_shared_ptr2 = B_loaded_fp16;
 
@@ -329,14 +433,10 @@ __global__ void __launch_bounds__(64) dequantize_weights(
 }  // namespace awq
 }  // namespace vllm
 
-torch::Tensor awq_dequantize(
+torch::Tensor awq_dequantize(torch::Tensor _kernel,
-    torch::Tensor _kernel,
                              torch::Tensor _scaling_factors,
-    torch::Tensor _zeros,
+                             torch::Tensor _zeros, int64_t split_k_iters,
-    int split_k_iters,
+                             int64_t thx, int64_t thy) {
-    int thx,
-    int thy)
-{
   int in_c = _kernel.size(0);
   int qout_c = _kernel.size(1);
   int out_c = qout_c * 8;
@@ -362,12 +462,15 @@ torch::Tensor awq_dequantize(
 
   const at::cuda::OptionalCUDAGuard device_guard(device_of(_scaling_factors));
 
-    auto options = torch::TensorOptions().dtype(_scaling_factors.dtype()).device(_scaling_factors.device());
+  auto options = torch::TensorOptions()
+                     .dtype(_scaling_factors.dtype())
+                     .device(_scaling_factors.device());
   at::Tensor _de_kernel = torch::empty({in_c, out_c}, options);
 
   auto kernel = reinterpret_cast<int*>(_kernel.data_ptr<int>());
   auto de_kernel = reinterpret_cast<half*>(_de_kernel.data_ptr<at::Half>());
-    auto scaling_factors = reinterpret_cast<half*>(_scaling_factors.data_ptr<at::Half>());
+  auto scaling_factors =
+      reinterpret_cast<half*>(_scaling_factors.data_ptr<at::Half>());
   auto zeros = reinterpret_cast<int*>(_zeros.data_ptr<int>());
 
   dim3 num_blocks(x_blocks, y_blocks);
@@ -386,26 +489,26 @@ torch::Tensor awq_dequantize(
 // zeros: IC // G, OC // 8 [int32] -> cast to IC // G, OC [uint4b]
 // assume that batch_size < 16 for now
 
-torch::Tensor awq_gemm(
+torch::Tensor awq_gemm(torch::Tensor _in_feats, torch::Tensor _kernel,
-    torch::Tensor _in_feats,
+                       torch::Tensor _scaling_factors, torch::Tensor _zeros,
-    torch::Tensor _kernel,
+                       int64_t split_k_iters) {
-    torch::Tensor _scaling_factors,
-    torch::Tensor _zeros,
-    int split_k_iters)
-{
   int num_in_feats = _in_feats.size(0);
   int num_in_channels = _in_feats.size(1);
   const at::cuda::OptionalCUDAGuard device_guard(device_of(_in_feats));
 
-    auto options = torch::TensorOptions().dtype(_in_feats.dtype()).device(_in_feats.device());
+  auto options = torch::TensorOptions()
-    at::Tensor _out_feats = torch::empty({split_k_iters, num_in_feats, _kernel.size(1) * 8}, options);
+                     .dtype(_in_feats.dtype())
+                     .device(_in_feats.device());
+  at::Tensor _out_feats =
+      torch::empty({split_k_iters, num_in_feats, _kernel.size(1) * 8}, options);
   int num_out_feats = _out_feats.size(-2);
   int num_out_channels = _out_feats.size(-1);
 
   auto in_feats = reinterpret_cast<half*>(_in_feats.data_ptr<at::Half>());
   auto kernel = reinterpret_cast<int*>(_kernel.data_ptr<int>());
   auto out_feats = reinterpret_cast<half*>(_out_feats.data_ptr<at::Half>());
-    auto scaling_factors = reinterpret_cast<half*>(_scaling_factors.data_ptr<at::Half>());
+  auto scaling_factors =
+      reinterpret_cast<half*>(_scaling_factors.data_ptr<at::Half>());
   auto zeros = reinterpret_cast<int*>(_zeros.data_ptr<int>());
   int group_size = num_in_channels / _scaling_factors.size(0);
 
@@ -419,28 +522,28 @@ torch::Tensor awq_gemm(
     throw std::invalid_argument("OC is not multiple of Group size");
 
   const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
-    if (num_out_channels % 128 == 0)
+  if (num_out_channels % 128 == 0) {
-    {
     int j_factors1 = num_out_channels / 128 / 1;
     dim3 num_blocks((num_out_feats + 16 - 1) / 16 * j_factors1 * split_k_iters);
     // threadIdx.x: 32
     // threadIdx.y: i_factors[2] * j_factors[2]
     dim3 threads_per_block(32, 2);
-        vllm::awq::gemm_forward_4bit_cuda_m16nXk32<128><<<num_blocks, threads_per_block, 0, stream>>>(
+    vllm::awq::gemm_forward_4bit_cuda_m16nXk32<128>
-            group_size, split_k_iters, in_feats, kernel, scaling_factors, zeros, num_in_feats, num_in_channels,
+        <<<num_blocks, threads_per_block, 0, stream>>>(
-            num_out_channels, out_feats);
+            group_size, split_k_iters, in_feats, kernel, scaling_factors, zeros,
-    }
+            num_in_feats, num_in_channels, num_out_channels, out_feats);
-    else if (num_out_channels % 64 == 0)
+  } else if (num_out_channels % 64 == 0) {
-    {
     int j_factors1 = num_out_channels / 64 / 1;
-        dim3 num_blocks(1 * (num_out_feats + 16 - 1) / 16 * j_factors1 * split_k_iters);
+    dim3 num_blocks(1 * (num_out_feats + 16 - 1) / 16 * j_factors1 *
+                    split_k_iters);
 
     // threadIdx.x: 32
     // threadIdx.y: i_factors[2] * j_factors[2]
     dim3 threads_per_block(32, 2);
-        vllm::awq::gemm_forward_4bit_cuda_m16nXk32<64><<<num_blocks, threads_per_block, 0, stream>>>(
+    vllm::awq::gemm_forward_4bit_cuda_m16nXk32<64>
-            group_size, split_k_iters, in_feats, kernel, scaling_factors, zeros, num_in_feats, num_in_channels,
+        <<<num_blocks, threads_per_block, 0, stream>>>(
-            num_out_channels, out_feats);
+            group_size, split_k_iters, in_feats, kernel, scaling_factors, zeros,
+            num_in_feats, num_in_channels, num_out_channels, out_feats);
   }
   return _out_feats.sum(0);
 }

csrc/quantization/compressed_tensors/int8_quant_kernels.cu

@@ -0,0 +1,115 @@
+#include <ATen/cuda/CUDAContext.h>
+#include <torch/all.h>
+#include <cmath>
+
+#include "../../dispatch_utils.h"
+#include "../../reduction_utils.cuh"
+
+static inline __device__ int8_t float_to_int8_rn(float x) {
+#ifdef USE_ROCM
+  static const float i8_min =
+      static_cast<float>(std::numeric_limits<int8_t>::min());
+  static const float i8_max =
+      static_cast<float>(std::numeric_limits<int8_t>::max());
+  // round
+  float dst = std::nearbyint(x);
+  // saturate
+  dst = std::clamp(dst, i8_min, i8_max);
+  return static_cast<int8_t>(dst);
+#else
+  // CUDA path
+  uint32_t dst;
+  asm volatile("cvt.rni.sat.s8.f32 %0, %1;" : "=r"(dst) : "f"(x));
+  return reinterpret_cast<const int8_t&>(dst);
+#endif
+}
+
+namespace vllm {
+
+template <typename scalar_t, typename scale_type>
+__global__ void static_scaled_int8_quant_kernel(
+    scalar_t const* __restrict__ input, int8_t* __restrict__ out,
+    scale_type const* scale_ptr, const int hidden_size) {
+  int const tid = threadIdx.x;
+  int const token_idx = blockIdx.x;
+  scale_type const scale = *scale_ptr;
+
+  for (int i = tid; i < hidden_size; i += blockDim.x) {
+    out[token_idx * hidden_size + i] = float_to_int8_rn(
+        static_cast<float>(input[token_idx * hidden_size + i]) / scale);
+  }
+}
+
+template <typename scalar_t, typename scale_type>
+__global__ void dynamic_scaled_int8_quant_kernel(
+    scalar_t const* __restrict__ input, int8_t* __restrict__ out,
+    scale_type* scale, const int hidden_size) {
+  int const tid = threadIdx.x;
+  int const token_idx = blockIdx.x;
+  float absmax_val = 0.0f;
+  float const zero = 0.0f;
+
+  for (int i = tid; i < hidden_size; i += blockDim.x) {
+    float val = static_cast<float>(input[token_idx * hidden_size + i]);
+    val = val > zero ? val : -val;
+    absmax_val = val > absmax_val ? val : absmax_val;
+  }
+
+  float const block_absmax_val_maybe = blockReduceMax(absmax_val);
+  __shared__ float block_absmax_val;
+  if (tid == 0) {
+    block_absmax_val = block_absmax_val_maybe;
+    scale[token_idx] = block_absmax_val / 127.0f;
+  }
+  __syncthreads();
+
+  float const tmp_scale = 127.0f / block_absmax_val;
+  for (int i = tid; i < hidden_size; i += blockDim.x) {
+    out[token_idx * hidden_size + i] = float_to_int8_rn(
+        static_cast<float>(input[token_idx * hidden_size + i]) * tmp_scale);
+  }
+}
+
+}  // namespace vllm
+
+void static_scaled_int8_quant(torch::Tensor& out,          // [..., hidden_size]
+                              torch::Tensor const& input,  // [..., hidden_size]
+                              torch::Tensor const& scale) {
+  TORCH_CHECK(input.is_contiguous());
+  TORCH_CHECK(out.is_contiguous());
+  TORCH_CHECK(scale.numel() == 1);
+
+  int const hidden_size = input.size(-1);
+  int const num_tokens = input.numel() / hidden_size;
+  dim3 const grid(num_tokens);
+  dim3 const block(std::min(hidden_size, 1024));
+  const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
+  VLLM_DISPATCH_FLOATING_TYPES(
+      input.scalar_type(), "static_scaled_int8_quant_kernel", [&] {
+        vllm::static_scaled_int8_quant_kernel<scalar_t, float>
+            <<<grid, block, 0, stream>>>(input.data_ptr<scalar_t>(),
+                                         out.data_ptr<int8_t>(),
+                                         scale.data_ptr<float>(), hidden_size);
+      });
+}
+
+void dynamic_scaled_int8_quant(
+    torch::Tensor& out,          // [..., hidden_size]
+    torch::Tensor const& input,  // [..., hidden_size]
+    torch::Tensor& scales) {
+  TORCH_CHECK(input.is_contiguous());
+  TORCH_CHECK(out.is_contiguous());
+
+  int const hidden_size = input.size(-1);
+  int const num_tokens = input.numel() / hidden_size;
+  dim3 const grid(num_tokens);
+  dim3 const block(std::min(hidden_size, 1024));
+  const cudaStream_t stream = at::cuda::getCurrentCUDAStream();
+  VLLM_DISPATCH_FLOATING_TYPES(
+      input.scalar_type(), "dynamic_scaled_int8_quant_kernel", [&] {
+        vllm::dynamic_scaled_int8_quant_kernel<scalar_t, float>
+            <<<grid, block, 0, stream>>>(input.data_ptr<scalar_t>(),
+                                         out.data_ptr<int8_t>(),
+                                         scales.data_ptr<float>(), hidden_size);
+      });
+}

csrc/quantization/cutlass_w8a8/broadcast_load_epilogue_c2x.hpp

@@ -0,0 +1,346 @@
+/***************************************************************************************************
+ * Copyright (c) 2023 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights
+ *reserved. SPDX-License-Identifier: BSD-3-Clause
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ * this list of conditions and the following disclaimer in the documentation
+ * and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ *ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ *LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ *CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ *SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ *INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ *CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ *ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ *POSSIBILITY OF SUCH DAMAGE.
+ *
+ **************************************************************************************************/
+
+//
+// This file is a modified excerpt of
+// include/cutlass/epilogue/fusion/visitor_load.hpp from
+// https://github.com/NVIDIA/cutlass v3.5.0
+// It has been modified to support either
+// row/column or scalar broadcasting where the tensor being loaded from is
+// always passed in via a device pointer. This lets one compiled kernel handle
+// all cases of per-tensor or per-channel/per-token quantization.
+//
+// This interface also allows the scales to be passed in as tensors that
+// consistently reside on the device, which avoids an issue with a previous
+// implementation where scalars needed to be on the CPU since they
+// were passed in via float values. This created a potential performance hazard
+// if scales were initially on the device, and caused torch.compile graph
+// breaks when moving scales to the CPU.
+//
+#pragma once
+
+// Turn off clang-format for the entire file to keep it close to upstream
+// clang-format off
+
+#include "cutlass/epilogue/threadblock/fusion/visitor_2x.hpp"
+#include "cute/tensor.hpp"
+
+namespace cutlass::epilogue::threadblock {
+
+using namespace cute;
+using namespace detail;
+
+template<
+  class ThreadMap,
+  class Element,
+  class StrideMNL
+>
+struct VisitorRowOrScalarBroadcast {
+
+  // This struct has been modified to have a bool indicating that ptr_row is a 
+  // scalar that must be broadcast.
+  struct Arguments {
+    Element const* ptr_row = nullptr;
+    bool row_broadcast = true;
+    StrideMNL dRow = {};
+  };
+
+  using Params = Arguments;
+
+  template <class ProblemShape>
+  static constexpr Params
+  to_underlying_arguments(ProblemShape const& problem_shape, Arguments const& args, void* workspace) {
+    return args;
+  }
+
+  template <class ProblemShape>
+  static size_t
+  get_workspace_size(ProblemShape const& problem_shape, Arguments const& args) {
+    return 0;
+  }
+
+  struct SharedStorage {};
+
+  // Global load type
+  static int constexpr vec_bits = ThreadMap::kElementsPerAccess * sizeof_bits<Element>::value;
+  using VecType = uint_bit_t<cute::min(128, vec_bits)>;
+  static int constexpr VecLength = sizeof(VecType) / sizeof(Element);
+
+  CUTLASS_HOST_DEVICE
+  VisitorRowOrScalarBroadcast() { }
+
+  CUTLASS_HOST_DEVICE
+  VisitorRowOrScalarBroadcast(Params const& params, SharedStorage const& shared_storage)
+    : params_ptr(&params) { }
+
+  Params const* params_ptr;
+
+  template <class GTensor, class RTensor, class CTensor, class ProblemShape>
+  struct Callbacks : EmptyCallbacks {
+    CUTLASS_DEVICE
+    Callbacks(
+      GTensor&& tC_gRow,
+      RTensor&& tC_rRow,
+      CTensor&& tC_cRow,
+      ProblemShape problem_shape,
+      Params const* params_ptr
+    ):
+      tC_gRow(cute::forward<GTensor>(tC_gRow)),
+      tC_rRow(cute::forward<RTensor>(tC_rRow)),
+      tC_cRow(cute::forward<CTensor>(tC_cRow)),
+      n(get<1>(problem_shape)),
+      params_ptr(params_ptr) { }
+
+    GTensor tC_gRow;
+    RTensor tC_rRow;
+    CTensor tC_cRow;
+    Params const* params_ptr;
+    int n;
+
+    // This function is modified from VisitorRowBroadcast
+    CUTLASS_DEVICE void
+    begin_epilogue() {
+      clear(tC_rRow);
+      auto src_v = filter(tC_gRow);
+      auto coord_v = filter(tC_cRow);
+      auto dst_v = filter(tC_rRow);
+
+      if (params_ptr->row_broadcast) {
+        // In this case we are loading from a row vector and broadcasting
+        CUTLASS_PRAGMA_UNROLL
+        for (int i = 0; i < size(src_v); ++i) {
+          bool guard = get<1>(coord_v(i)) < n;
+          cutlass::arch::global_load<VecType, sizeof(VecType)>(
+              dst_v(i), (void const*)&src_v(i), guard);
+        }
+      } else {
+        // In this case we are loading from a scalar and broadcasting
+        VecType filled_vec;
+        CUTLASS_PRAGMA_UNROLL
+        for (int i = 0; i < VecLength; i++) {
+          reinterpret_cast<Element*>(&filled_vec)[i] = *(params_ptr->ptr_row);
+        }
+
+        CUTLASS_PRAGMA_UNROLL
+        for (int i = 0; i < size(src_v); ++i) {
+          if (get<1>(coord_v(i)) < n) {
+            dst_v(i) = filled_vec;
+          }
+        }
+      }
+    }
+
+    template <class ElementAccumulator, int FragmentSize>
+    CUTLASS_DEVICE auto // returns an Array
+    visit(int iter_idx, int row_idx, int column_idx, int frg_idx,
+          Array<ElementAccumulator, FragmentSize> const& frg_acc) {
+      Tensor rRow_frg = recast<Array<Element, FragmentSize>>(coalesce(tC_rRow));
+      return rRow_frg(column_idx);
+    }
+  };
+
+  template <class ProblemShape>
+  CUTLASS_DEVICE auto
+  get_callbacks(
+    gemm::GemmCoord threadblock_tile_offset,
+    int thread_idx,
+    ProblemShape problem_shape
+  ) {
+    Tensor mRow = make_tensor(
+      make_gmem_ptr(params_ptr->ptr_row),
+      problem_shape,
+      params_ptr->dRow);
+
+    // VECTOR, FRAGMENT_COLUMN
+    Tensor tC_gRow = recast<VecType>(
+      ThreadMap::partition(mRow, thread_idx, threadblock_tile_offset)
+    )(_,_,_0{},_0{},_0{},_0{});
+    Tensor tC_rRow = make_tensor_like(tC_gRow);
+
+    // Generate the pred tensor
+    Tensor cRow = make_identity_tensor(mRow.shape());
+    Tensor tC_cRow = outer_partition(
+      ThreadMap::partition(cRow, thread_idx, threadblock_tile_offset)(_,_,_0{},_0{},_0{},_0{}),
+      Shape<Int<VecLength>>{},
+      (_0{})
+    );
+
+    return Callbacks<
+      decltype(tC_gRow), decltype(tC_rRow),
+      decltype(tC_cRow), ProblemShape>(
+      cute::move(tC_gRow),
+      cute::move(tC_rRow),
+      cute::move(tC_cRow),
+      problem_shape,
+      params_ptr
+    );
+  }
+
+};
+
+/////////////////////////////////////////////////////////////////////////////////////////////////
+
+// Column vector broadcast
+template<
+  class ThreadMap,
+  class Element,
+  class StrideMNL = Stride<_1,_0,_0>
+>
+struct VisitorColOrScalarBroadcast {
+
+  // This struct has been modified to have a bool indicating that ptr_col is a 
+  // scalar that must be broadcast.
+  struct Arguments {
+    Element const* ptr_col = nullptr;
+    bool col_broadcast = true;
+    StrideMNL dCol = {};
+  };
+
+  using Params = Arguments;
+
+  template <class ProblemShape>
+  static constexpr Params
+  to_underlying_arguments(ProblemShape const& problem_shape, Arguments const& args, void* workspace) {
+    return args;
+  }
+
+  template <class ProblemShape>
+  static size_t
+  get_workspace_size(ProblemShape const& problem_shape, Arguments const& args) {
+    return 0;
+  }
+
+  struct SharedStorage { };
+
+  CUTLASS_HOST_DEVICE
+  VisitorColOrScalarBroadcast() { }
+
+  CUTLASS_HOST_DEVICE
+  VisitorColOrScalarBroadcast(Params const& params, SharedStorage const& shared_storage)
+    : params_ptr(&params) { }
+
+  Params const* params_ptr;
+
+  template <class GTensor, class RTensor, class CTensor, class ProblemShape>
+  struct Callbacks : EmptyCallbacks {
+    CUTLASS_DEVICE
+    Callbacks(
+      GTensor&& tC_gCol,
+      RTensor&& tC_rCol,
+      CTensor&& tC_cCol,
+      ProblemShape problem_shape,
+      Params const* params_ptr
+    ):
+      tC_gCol(cute::forward<GTensor>(tC_gCol)),
+      tC_rCol(cute::forward<RTensor>(tC_rCol)),
+      tC_cCol(cute::forward<CTensor>(tC_cCol)),
+      m(get<0>(problem_shape)),
+      params_ptr(params_ptr) { }
+
+    GTensor tC_gCol;
+    RTensor tC_rCol;
+    CTensor tC_cCol;
+    Params const* params_ptr;
+    int m;
+
+    // This function is modified from VisitorColBroadcast
+    CUTLASS_DEVICE void 
+    begin_epilogue() {
+      clear(tC_rCol);
+
+      Tensor pred = make_tensor<bool>(shape(tC_gCol));
+      CUTLASS_PRAGMA_UNROLL
+      for (int i = 0; i < size(pred); ++i) {
+        pred(i) = get<0>(tC_cCol(i)) < m;
+      }
+
+      if (params_ptr->col_broadcast) {
+        // In this case we are loading from a column vector and broadcasting
+        copy_if(pred, tC_gCol, tC_rCol);
+      } else {
+        // In this case we are loading from a scalar and broadcasting
+        auto dst_v = filter(tC_rCol);
+
+        CUTLASS_PRAGMA_UNROLL
+        for (int i = 0; i < size(dst_v); ++i) {
+          if (pred(i)) {
+            dst_v(i) = *(params_ptr->ptr_col);
+          }
+        }
+      }
+    }
+
+    template <class ElementAccumulator, int FragmentSize>
+    CUTLASS_DEVICE auto // returns an Array
+    visit(int iter_idx, int row_idx, int column_idx, int frg_idx,
+          Array<ElementAccumulator, FragmentSize> const& frg_acc) {
+      Array<Element, FragmentSize> frg_col;
+      frg_col.fill(tC_rCol(row_idx,iter_idx));
+      return frg_col;
+    }
+  };
+
+  template <class ProblemShape>
+  CUTLASS_DEVICE auto
+  get_callbacks(
+    gemm::GemmCoord threadblock_tile_offset,
+    int thread_idx,
+    ProblemShape problem_shape
+  ) {
+    Tensor mCol = make_tensor(
+      make_gmem_ptr(params_ptr->ptr_col),
+      problem_shape,
+      params_ptr->dCol);
+
+    // VECTOR, FRAGMENT_COLUMN, FRAGMENT_ROW, ITERATION_ROW, ITERATION_GROUP, ITERATION_CLUSTER
+    Tensor tC_gCol = group_modes<1,4>(
+      ThreadMap::partition(mCol, thread_idx, threadblock_tile_offset)(_0{},_0{},_,_,_,_));
+    Tensor tC_rCol = make_tensor_like(tC_gCol);
+
+    // Generate the pred tensor
+    Tensor cCol = make_identity_tensor(mCol.shape());
+    Tensor tC_cCol = group_modes<1,4>(
+      ThreadMap::partition(cCol, thread_idx, threadblock_tile_offset)(_0{},_0{},_,_,_,_));
+
+    return Callbacks<
+      decltype(tC_gCol), decltype(tC_rCol),
+      decltype(tC_cCol), ProblemShape>(
+      cute::move(tC_gCol),
+      cute::move(tC_rCol),
+      cute::move(tC_cCol),
+      problem_shape,
+      params_ptr
+    );
+  }
+};
+
+}

csrc/quantization/cutlass_w8a8/broadcast_load_epilogue_c3x.hpp

@@ -0,0 +1,389 @@
+/***************************************************************************************************
+ * Copyright (c) 2023 - 2024 NVIDIA CORPORATION & AFFILIATES. All rights
+ *reserved. SPDX-License-Identifier: BSD-3-Clause
+ *
+ * Redistribution and use in source and binary forms, with or without
+ * modification, are permitted provided that the following conditions are met:
+ *
+ * 1. Redistributions of source code must retain the above copyright notice,
+ *this list of conditions and the following disclaimer.
+ *
+ * 2. Redistributions in binary form must reproduce the above copyright notice,
+ * this list of conditions and the following disclaimer in the documentation
+ * and/or other materials provided with the distribution.
+ *
+ * 3. Neither the name of the copyright holder nor the names of its
+ * contributors may be used to endorse or promote products derived from
+ * this software without specific prior written permission.
+ *
+ * THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS "AS IS"
+ * AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
+ * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
+ *ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT HOLDER OR CONTRIBUTORS BE
+ *LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
+ *CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
+ *SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
+ *INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
+ *CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
+ *ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
+ *POSSIBILITY OF SUCH DAMAGE.
+ *
+ **************************************************************************************************/
+
+//
+// This file is a modified excerpt of
+// include/cutlass/epilogue/fusion/sm90_visitor_load_tma_warpspecialized.hpp
+// from https://github.com/NVIDIA/cutlass v3.5.0
+// It has been modified to support either row/column or scalar broadcasting
+// where the tensor being loaded from is always passed in via a device pointer.
+// This lets one compiled kernel handle all cases of per-tensor or
+// per-channel/per-token quantization.
+//
+// This interface also allows the scales to be passed in as tensors that
+// consistently reside on the device, which avoids an issue with a previous
+// implementation where scalars needed to be on the CPU since they
+// were passed in via float values. This created a potential performance hazard
+// if scales were initially on the device, and caused torch.compile graphs
+// breaks when moving scales to the CPU.
+//
+#pragma once
+
+// Turn off clang-format for the entire file to keep it close to upstream
+// clang-format off
+
+#include "cutlass/cutlass.h"
+#include "cutlass/arch/barrier.h"
+
+#include "cute/tensor.hpp"
+#include "cutlass/epilogue/fusion/sm90_visitor_tma_warpspecialized.hpp"
+
+namespace cutlass::epilogue::fusion {
+
+using namespace cute;
+using namespace detail;
+
+// Row vector broadcast
+template<
+  // Row bcast reuses the mbarriers from the epilogue subtile load pipeline, so this must be at least
+  // ceil_div(StagesC, epi tiles per CTA tile) + 1 to ensure no data races
+  int Stages,
+  class CtaTileShapeMNK,
+  class Element,
+  class StrideMNL = Stride<_0,_1,_0>,
+  int Alignment = 128 / sizeof_bits_v<Element>
+>
+struct Sm90RowOrScalarBroadcast {
+  static_assert(Alignment * sizeof_bits_v<Element> % 128 == 0, "sub-16B alignment not supported yet");
+  static_assert(
+    (cute::is_same_v<StrideMNL, Stride<_0,_1, _0>>) || // row vector broadcast, e.g. per-col alpha/bias
+    (cute::is_same_v<StrideMNL, Stride<_0,_1,int>>));  // batched row vector broadcast
+
+  // Accumulator doesn't distribute row elements evenly amongst threads so we must buffer in smem
+  struct SharedStorage {
+    alignas(16) array_aligned<Element, size<1>(CtaTileShapeMNK{}) * Stages> smem_row;
+  };
+
+  // This struct has been modified to have a bool indicating that ptr_row is a 
+  // scalar that must be broadcast, instead of containing a scalar that is 
+  // valid if ptr_row is null.
+  struct Arguments {
+    Element const* ptr_row = nullptr;
+    bool row_broadcast = true;
+    StrideMNL dRow = {};
+  };
+
+  using Params = Arguments;
+
+  template <class ProblemShape>
+  static constexpr Params
+  to_underlying_arguments(ProblemShape const& problem_shape, Arguments const& args, void* workspace) {
+    return args;
+  }
+
+  template <class ProblemShape>
+  static size_t
+  get_workspace_size(ProblemShape const& problem_shape, Arguments const& args) {
+    return 0;
+  }
+
+  template <class ProblemShape>
+  static cutlass::Status
+  initialize_workspace(ProblemShape const& problem_shape, Arguments const& args, void* workspace, cudaStream_t stream,
+    CudaHostAdapter* cuda_adapter = nullptr) {
+    return cutlass::Status::kSuccess;
+  }
+
+  CUTLASS_HOST_DEVICE
+  Sm90RowOrScalarBroadcast() { }
+
+  CUTLASS_HOST_DEVICE
+  Sm90RowOrScalarBroadcast(Params const& params, SharedStorage const& shared_storage)
+      : params(params),
+        smem_row(const_cast<Element*>(shared_storage.smem_row.data())) { }
+
+  Params params;
+  Element* smem_row;
+
+  CUTLASS_DEVICE bool
+  is_producer_load_needed() const {
+    return true;
+  }
+
+  CUTLASS_DEVICE bool
+  is_C_load_needed() const {
+    return false;
+  }
+
+  CUTLASS_DEVICE bool
+  is_zero() const {
+    return (!params.row_broadcast && *(params.ptr_row) == Element(0));
+  }
+
+  template <int EpiTiles, class GTensor, class STensor>
+  struct ProducerLoadCallbacks : EmptyProducerLoadCallbacks {
+    CUTLASS_DEVICE
+    ProducerLoadCallbacks(GTensor&& gRow, STensor&& sRow, Params const& params)
+      : gRow(cute::forward<GTensor>(gRow)),
+        sRow(cute::forward<STensor>(sRow)),
+        params(params) {}
+
+    GTensor gRow;                                                                                 // (CTA_M,CTA_N)
+    STensor sRow;                                                                                 // (CTA_M,CTA_N,PIPE)
+    Params const& params;
+
+    CUTLASS_DEVICE void
+    begin(uint64_t* full_mbarrier_ptr, int load_iteration, bool issue_tma_load) {
+      if (params.ptr_row == nullptr) {
+        return;
+      }
+
+      if (issue_tma_load) {
+        // Increment the expect-tx count of the first subtile's mbarrier by the row vector's byte-size
+        constexpr uint32_t copy_bytes = size<1>(CtaTileShapeMNK{}) * sizeof_bits_v<Element> / 8;
+        cutlass::arch::ClusterTransactionBarrier::expect_transaction(full_mbarrier_ptr, copy_bytes);
+        // Issue the TMA bulk copy
+        auto bulk_copy = Copy_Atom<SM90_BULK_COPY_AUTO, Element>{}.with(*full_mbarrier_ptr);
+        // Filter so we don't issue redundant copies over stride-0 modes
+        int bcast_pipe_index = (load_iteration / EpiTiles) % Stages;
+        copy(bulk_copy, filter(gRow), filter(sRow(_,_,bcast_pipe_index)));
+      }
+    }
+  };
+
+  template <class... Args>
+  CUTLASS_DEVICE auto
+  get_producer_load_callbacks(ProducerLoadArgs<Args...> const& args) {
+
+    auto [M, N, K, L] = args.problem_shape_mnkl;
+    auto [m, n, k, l] = args.tile_coord_mnkl;
+    Tensor mRow = make_tensor(make_gmem_ptr(params.ptr_row), make_shape(M,N,L), params.dRow);
+    Tensor gRow = local_tile(mRow, take<0,2>(args.tile_shape_mnk), make_coord(m,n,l));            // (CTA_M,CTA_N)
+    Tensor sRow = make_tensor(make_smem_ptr(smem_row),                                            // (CTA_M,CTA_N,PIPE)
+                    make_shape(size<0>(CtaTileShapeMNK{}), size<1>(CtaTileShapeMNK{}), Stages),
+                    make_stride(_0{},_1{},size<1>(CtaTileShapeMNK{})));
+
+    constexpr int EpiTiles = decltype(size<1>(zipped_divide(make_layout(take<0,2>(args.tile_shape_mnk)), args.epi_tile)))::value;
+    return ProducerLoadCallbacks<EpiTiles, decltype(gRow), decltype(sRow)>(
+      cute::move(gRow), cute::move(sRow), params);
+  }
+
+  template <int EpiTiles, class RTensor, class STensor>
+  struct ConsumerStoreCallbacks : EmptyConsumerStoreCallbacks {
+    CUTLASS_DEVICE
+    ConsumerStoreCallbacks(RTensor&& tCrRow, STensor&& tCsRow, Params const& params)
+      : tCrRow(cute::forward<RTensor>(tCrRow)),
+        tCsRow(cute::forward<STensor>(tCsRow)),
+        params(params) {}
+
+    RTensor tCrRow;                                                               // (CPY,CPY_M,CPY_N)
+    STensor tCsRow;                                                               // (CPY,CPY_M,CPY_N,EPI_M,EPI_N,PIPE)
+    Params const& params;
+
+    CUTLASS_DEVICE void
+    previsit(int epi_m, int epi_n, int load_iteration, bool is_producer_load_needed) {
+      if (!params.row_broadcast) {
+        fill(tCrRow, *(params.ptr_row));
+        return;
+      }
+
+      if (epi_m == 0) { // Assumes M-major subtile loop
+        // Filter so we don't issue redundant copies over stride-0 modes
+        // (only works if 0-strides are in same location, which is by construction)
+        int bcast_pipe_index = (load_iteration / EpiTiles) % Stages;
+        copy_aligned(filter(tCsRow(_,_,_,epi_m,epi_n,bcast_pipe_index)), filter(tCrRow));
+      }
+    }
+
+    template <typename ElementAccumulator, int FragmentSize>
+    CUTLASS_DEVICE Array<Element, FragmentSize>
+    visit(Array<ElementAccumulator, FragmentSize> const& frg_acc, int epi_v, int epi_m, int epi_n) {
+      Array<Element, FragmentSize> frg_row;
+
+      CUTLASS_PRAGMA_UNROLL
+      for (int i = 0; i < FragmentSize; ++i) {
+        frg_row[i] = tCrRow(epi_v * FragmentSize + i);
+      }
+
+      return frg_row;
+    }
+  };
+
+  template <
+    bool ReferenceSrc, // do register tensors reference the src or dst layout of the tiled copy
+    class... Args
+  >
+  CUTLASS_DEVICE auto
+  get_consumer_store_callbacks(ConsumerStoreArgs<Args...> const& args) {
+
+    Tensor sRow = make_tensor(make_smem_ptr(smem_row),                                            // (CTA_M,CTA_N,PIPE)
+                    make_shape(size<0>(CtaTileShapeMNK{}), size<1>(CtaTileShapeMNK{}), Stages),
+                    make_stride(_0{},_1{},size<1>(CtaTileShapeMNK{})));
+    Tensor tCsRow = sm90_partition_for_epilogue<ReferenceSrc>(                    // (CPY,CPY_M,CPY_N,EPI_M,EPI_N,PIPE)
+                      sRow, args.epi_tile, args.tiled_copy, args.thread_idx);
+    Tensor tCrRow = make_tensor_like(take<0,3>(tCsRow));                                           // (CPY,CPY_M,CPY_N)
+
+    constexpr int EpiTiles = decltype(size<1>(zipped_divide(make_layout(take<0,2>(args.tile_shape_mnk)), args.epi_tile)))::value;
+    return ConsumerStoreCallbacks<EpiTiles, decltype(tCrRow), decltype(tCsRow)>(
+      cute::move(tCrRow), cute::move(tCsRow), params);
+  }
+};
+
+/////////////////////////////////////////////////////////////////////////////////////////////////
+
+// Column vector broadcast
+template<
+  int Stages,
+  class CtaTileShapeMNK,
+  class Element,
+  class StrideMNL = Stride<_1,_0,_0>,
+  int Alignment = 128 / sizeof_bits_v<Element>
+>
+struct Sm90ColOrScalarBroadcast {
+  static_assert(Stages == 0, "Column broadcast doesn't support smem usage yet");
+  static_assert(Alignment * sizeof_bits_v<Element> % 128 == 0, "sub-16B alignment not supported yet");
+  static_assert(
+    (cute::is_same_v<StrideMNL, Stride<_1,_0, _0>>) || // col vector broadcast, e.g. per-row alpha/bias
+    (cute::is_same_v<StrideMNL, Stride<_1,_0,int>>));  // batched col vector broadcast, e.g. batched per-row bias
+
+  // Accumulator distributes col elements evenly amongst threads so we can just directly load from gmem
+  struct SharedStorage { };
+
+  // This struct has been modified to have a bool indicating that ptr_col is a 
+  // scalar that must be broadcast, instead of containing a scalar that is 
+  // valid if ptr_col is null.
+  struct Arguments {
+    Element const* ptr_col = nullptr;
+    bool col_broadcast = true;
+    StrideMNL dCol = {};
+  };
+
+  using Params = Arguments;
+
+  template <class ProblemShape>
+  static constexpr Params
+  to_underlying_arguments(ProblemShape const& problem_shape, Arguments const& args, void* workspace) {
+    return args;
+  }
+
+  template <class ProblemShape>
+  static size_t
+  get_workspace_size(ProblemShape const& problem_shape, Arguments const& args) {
+    return 0;
+  }
+
+  template <class ProblemShape>
+  static cutlass::Status
+  initialize_workspace(ProblemShape const& problem_shape, Arguments const& args, void* workspace, cudaStream_t stream,
+    CudaHostAdapter* cuda_adapter = nullptr) {
+    return cutlass::Status::kSuccess;
+  }
+
+  CUTLASS_DEVICE bool
+  is_producer_load_needed() const {
+    return false;
+  }
+
+  CUTLASS_DEVICE bool
+  is_C_load_needed() const {
+    return false;
+  }
+
+  CUTLASS_DEVICE bool
+  is_zero() const {
+    return (!params.col_broadcast && *(params.ptr_col) == Element(0));
+  }
+
+  CUTLASS_HOST_DEVICE
+  Sm90ColOrScalarBroadcast() { }
+
+  CUTLASS_HOST_DEVICE
+  Sm90ColOrScalarBroadcast(Params const& params, SharedStorage const& shared_storage)
+      : params(params) { }
+
+  Params params;
+
+  template <class... Args>
+  CUTLASS_DEVICE auto
+  get_producer_load_callbacks(ProducerLoadArgs<Args...> const& args) {
+    return EmptyProducerLoadCallbacks{};
+  }
+
+  template<class GTensor, class RTensor>
+  struct ConsumerStoreCallbacks : EmptyConsumerStoreCallbacks {
+    CUTLASS_DEVICE
+    ConsumerStoreCallbacks(GTensor&& tCgCol, RTensor&& tCrCol, Params const& params)
+      : tCgCol(cute::forward<GTensor>(tCgCol)),
+        tCrCol(cute::forward<RTensor>(tCrCol)),
+        params(params) {}
+
+    GTensor tCgCol;                                                                    // (CPY,CPY_M,CPY_N,EPI_M,EPI_N)
+    RTensor tCrCol;                                                                    // (CPY,CPY_M,CPY_N,EPI_M,EPI_N)
+    Params const& params;
+
+    CUTLASS_DEVICE void
+    begin() {
+      if (!params.col_broadcast) {
+        fill(tCrCol, *(params.ptr_col));
+        return;
+      }
+
+      // Filter so we don't issue redundant copies over stride-0 modes
+      // (only works if 0-strides are in same location, which is by construction)
+      copy_aligned(filter(tCgCol), filter(tCrCol));
+    }
+
+    template <typename ElementAccumulator, int FragmentSize>
+    CUTLASS_DEVICE Array<Element, FragmentSize>
+    visit(Array<ElementAccumulator, FragmentSize> const& frg_acc, int epi_v, int epi_m, int epi_n) {
+      Array<Element, FragmentSize> frg_col;
+      Tensor tCrCol_mn = tCrCol(_,_,_,epi_m,epi_n);
+
+      CUTLASS_PRAGMA_UNROLL
+      for (int i = 0; i < FragmentSize; ++i) {
+        frg_col[i] = tCrCol_mn(epi_v * FragmentSize + i);
+      }
+
+      return frg_col;
+    }
+
+  };
+
+  template <
+    bool ReferenceSrc, // do register tensors reference the src or dst layout of the tiled copy
+    class... Args
+  >
+  CUTLASS_DEVICE auto
+  get_consumer_store_callbacks(ConsumerStoreArgs<Args...> const& args) {
+
+    auto [M, N, K, L] = args.problem_shape_mnkl;
+    Tensor mCol = make_tensor(make_gmem_ptr(params.ptr_col), make_shape(M,N,L), params.dCol);
+    Tensor tCgCol = sm90_partition_for_epilogue<ReferenceSrc>(                         // (CPY,CPY_M,CPY_N,EPI_M,EPI_N)
+      mCol, args.tile_shape_mnk, args.tile_coord_mnkl, args.epi_tile, args.tiled_copy, args.thread_idx);
+    Tensor tCrCol = make_tensor_like(tCgCol);                                          // (CPY,CPY_M,CPY_N,EPI_M,EPI_N)
+
+    return ConsumerStoreCallbacks<decltype(tCgCol), decltype(tCrCol)>(
+      cute::move(tCgCol), cute::move(tCrCol), params);
+  }
+};
+
+}

csrc/quantization/cutlass_w8a8/common.hpp

@@ -0,0 +1,12 @@
+#pragma once
+
+#include "cutlass/cutlass.h"
+
+/**
+ * Helper function for checking CUTLASS errors
+ */
+#define CUTLASS_CHECK(status)                        \
+  {                                                  \
+    TORCH_CHECK(status == cutlass::Status::kSuccess, \
+                cutlassGetStatusString(status))      \
+  }

csrc/quantization/cutlass_w8a8/scaled_mm_dq_c2x.cu

@@ -0,0 +1,329 @@
+#include <stddef.h>
+#include <torch/all.h>
+
+#include <ATen/cuda/CUDAContext.h>
+
+// clang-format will break include orders
+// clang-format off
+#include "cute/tensor.hpp"
+#include "cute/atom/mma_atom.hpp"
+#include "cutlass/numeric_types.h"
+
+#include "cutlass/util/device_memory.h"
+
+#include "cutlass/cutlass.h"
+#include "cutlass/gemm_coord.h"
+#include "cutlass/arch/mma_sm75.h"
+#include "cutlass/arch/arch.h"
+#include "cutlass/arch/mma.h"
+#include "cutlass/gemm/device/gemm.h"
+#include "cutlass/gemm/device/gemm_universal_adapter.h"
+
+#include "cutlass/epilogue/threadblock/fusion/visitors.hpp"
+#include "cutlass/gemm/kernel/default_gemm_universal_with_visitor.h"
+
+#include "broadcast_load_epilogue_c2x.hpp"
+#include "common.hpp"
+// clang-format on
+
+using namespace cute;
+
+/*
+   This defines a quantized GEMM operation with dequantized output, similar to
+   torch._scaled_mm. It is defined using the CUTLASS 2.x API, and is used for
+   NVIDIA GPUs with SM versions prior to sm90 (Hopper).
+
+   A and B may be both either int8 or fp8_e4m3. A can be quantized per-tensor or
+   per-row. B can be quantized per-tensor or per-column.
+   Any combination of per-tensor and per-row or column is supported.
+   A and B must have symmetric quantization (zero point == 0).
+
+   So the GEMM operation is D = (a_scales * A) (b_scales * B), where the
+   scales are applied elementwise with numpy-style broadcasting.
+
+   ScaleA and ScaleB define the epilogue functions that apply the scales for
+   the A and B operands respectively. These scales may be either per-tensor or
+   per row or column.
+*/
+
+namespace {
+
+// Wrappers for the GEMM kernel that is used to guard against compilation on
+// architectures that will never use the kernel. The purpose of this is to
+// reduce the size of the compiled binary.
+// __CUDA_ARCH__ is not defined in host code, so this lets us smuggle the ifdef
+// into code that will be executed on the device where it is defined.
+template <typename Kernel>
+struct enable_sm75_to_sm80 : Kernel {
+  template <typename... Args>
+  CUTLASS_DEVICE static void invoke(Args&&... args) {
+#if defined __CUDA_ARCH__ && __CUDA_ARCH__ >= 750 && __CUDA_ARCH__ < 800
+    Kernel::invoke(std::forward<Args>(args)...);
+#endif
+  }
+};
+
+template <typename Kernel>
+struct enable_sm80_to_sm89 : Kernel {
+  template <typename... Args>
+  CUTLASS_DEVICE static void invoke(Args&&... args) {
+#if defined __CUDA_ARCH__ && __CUDA_ARCH__ >= 800 && __CUDA_ARCH__ < 890
+    Kernel::invoke(std::forward<Args>(args)...);
+#endif
+  }
+};
+
+template <typename Kernel>
+struct enable_sm89_to_sm90 : Kernel {
+  template <typename... Args>
+  CUTLASS_DEVICE static void invoke(Args&&... args) {
+#if defined __CUDA_ARCH__ && __CUDA_ARCH__ >= 890 && __CUDA_ARCH__ < 900
+    Kernel::invoke(std::forward<Args>(args)...);
+#endif
+  }
+};
+
+template <typename Arch, template <typename> typename ArchGuard,
+          typename ElementAB_, typename ElementD_, typename TileShape,
+          typename WarpShape, typename InstructionShape, int32_t MainLoopStages>
+struct cutlass_2x_gemm {
+  using ElementAB = ElementAB_;
+  using ElementD = ElementD_;
+
+  using ElementAcc =
+      typename std::conditional<std::is_same_v<ElementAB, int8_t>, int32_t,
+                                float>::type;
+
+  using Operator =
+      typename std::conditional<std::is_same_v<ElementAB, int8_t>,
+                                cutlass::arch::OpMultiplyAddSaturate,
+                                cutlass::arch::OpMultiplyAdd>::type;
+
+  using OutputTileThreadMap =
+      cutlass::epilogue::threadblock::OutputTileThreadLayout<
+          TileShape, WarpShape, float, 4, 1 /* epilogue stages */
+          >;
+
+  using Accum = cutlass::epilogue::threadblock::VisitorAccFetch;
+
+  using ScaleA = cutlass::epilogue::threadblock::VisitorColOrScalarBroadcast<
+      OutputTileThreadMap, float, Stride<Int<1>, Int<0>, Int<0>>>;
+
+  using ScaleB = cutlass::epilogue::threadblock::VisitorRowOrScalarBroadcast<
+      OutputTileThreadMap, float, Stride<Int<0>, Int<1>, Int<0>>>;
+
+  using Compute0 = cutlass::epilogue::threadblock::VisitorCompute<
+      cutlass::multiplies, float, float,
+      cutlass::FloatRoundStyle::round_to_nearest>;
+
+  using EVTCompute0 =
+      cutlass::epilogue::threadblock::Sm80EVT<Compute0, ScaleB, Accum>;
+
+  using Compute1 = cutlass::epilogue::threadblock::VisitorCompute<
+      cutlass::multiplies, ElementD, float,
+      cutlass::FloatRoundStyle::round_to_nearest>;
+
+  using EVTCompute1 =
+      cutlass::epilogue::threadblock::Sm80EVT<Compute1, ScaleA, EVTCompute0>;
+
+  using D = cutlass::epilogue::threadblock::VisitorAuxStore<
+      OutputTileThreadMap, ElementD, cutlass::FloatRoundStyle::round_to_nearest,
+      Stride<int64_t, Int<1>, Int<0>>>;
+
+  using EVTD = cutlass::epilogue::threadblock::Sm80EVT<D, EVTCompute1>;
+
+  // clang-format off
+  using RowMajor = typename cutlass::layout::RowMajor;
+  using ColumnMajor = typename cutlass::layout::ColumnMajor;
+  using KernelType = 
+    ArchGuard<typename cutlass::gemm::kernel::DefaultGemmWithVisitor<
+      ElementAB, RowMajor, cutlass::ComplexTransform::kNone, 16, 
+      ElementAB, ColumnMajor, cutlass::ComplexTransform::kNone, 16, 
+      float, cutlass::layout::RowMajor, 4,
+      ElementAcc, float, cutlass::arch::OpClassTensorOp, 
+      Arch, 
+      TileShape, WarpShape, InstructionShape,
+      EVTD,
+      cutlass::gemm::threadblock::ThreadblockSwizzleStreamK,
+      MainLoopStages, Operator,
+      1 /* epilogue stages */
+      >::GemmKernel>;
+  // clang-format on
+
+  using Op = cutlass::gemm::device::GemmUniversalAdapter<KernelType>;
+};
+
+template <typename Gemm>
+void cutlass_scaled_mm_dq_dispatcher(torch::Tensor& out, torch::Tensor const& a,
+                                     torch::Tensor const& b,
+                                     torch::Tensor const& a_scales,
+                                     torch::Tensor const& b_scales) {
+  using ElementAB = typename Gemm::ElementAB;
+  using ElementD = typename Gemm::ElementD;
+
+  int32_t m = a.size(0);
+  int32_t n = b.size(1);
+  int32_t k = a.size(1);
+  cutlass::gemm::GemmCoord problem_size{m, n, k};
+
+  int64_t lda = a.stride(0);
+  int64_t ldb = b.stride(1);
+  int64_t ldc = out.stride(0);
+
+  using StrideC = Stride<int64_t, Int<1>, Int<0>>;
+  StrideC c_stride{ldc, Int<1>{}, Int<0>{}};
+
+  auto a_ptr = static_cast<ElementAB const*>(a.data_ptr());
+  auto b_ptr = static_cast<ElementAB const*>(b.data_ptr());
+  auto c_ptr = static_cast<ElementD*>(out.data_ptr());
+
+  auto a_scales_ptr = a_scales.data_ptr<float>();
+  auto b_scales_ptr = b_scales.data_ptr<float>();
+
+  using ScaleAArgs = typename Gemm::ScaleA::Arguments;
+  using ScaleBArgs = typename Gemm::ScaleB::Arguments;
+
+  ScaleBArgs b_args{b_scales.data_ptr<float>(), b_scales.numel() != 1, {}};
+  ScaleAArgs a_args{a_scales.data_ptr<float>(), a_scales.numel() != 1, {}};
+
+  typename Gemm::EVTCompute0::Arguments evt0_compute_args{b_args};
+
+  typename Gemm::EVTCompute1::Arguments evt1_compute_args{a_args,
+                                                          evt0_compute_args};
+  typename Gemm::D::Arguments d_args{c_ptr, c_stride};
+
+  typename Gemm::EVTD::Arguments epilogue_args{
+      evt1_compute_args,
+      d_args,
+  };
+
+  typename Gemm::Op::Arguments args{
+      cutlass::gemm::GemmUniversalMode::kGemmSplitKParallel,  // universal mode
+      problem_size,                                           // problem size
+      1,                                                      // batch count
+      epilogue_args,
+      a_ptr,
+      b_ptr,
+      nullptr,
+      nullptr,
+      0,
+      0,
+      0,
+      0,
+      lda,
+      ldb,
+      ldc,
+      ldc};
+
+  // Launch the CUTLASS GEMM kernel.
+  typename Gemm::Op gemm_op;
+  size_t workspace_size = gemm_op.get_workspace_size(args);
+  cutlass::device_memory::allocation<uint8_t> workspace(workspace_size);
+
+  auto stream = at::cuda::getCurrentCUDAStream(a.get_device());
+
+  CUTLASS_CHECK(gemm_op.can_implement(args));
+  cutlass::Status status = gemm_op(args, workspace.get(), stream);
+  CUTLASS_CHECK(status);
+}
+
+}  // namespace
+
+void cutlass_scaled_mm_dq_sm75(torch::Tensor& out, torch::Tensor const& a,
+                               torch::Tensor const& b,
+                               torch::Tensor const& a_scales,
+                               torch::Tensor const& b_scales) {
+  TORCH_CHECK(a.dtype() == torch::kInt8);
+  TORCH_CHECK(b.dtype() == torch::kInt8);
+  TORCH_CHECK(a_scales.dtype() == torch::kFloat32);
+  TORCH_CHECK(b_scales.dtype() == torch::kFloat32);
+
+  using TileShape = typename cutlass::gemm::GemmShape<128, 128, 64>;
+  using WarpShape = typename cutlass::gemm::GemmShape<64, 64, 64>;
+  using InstructionShape = typename cutlass::gemm::GemmShape<8, 8, 16>;
+
+  if (out.dtype() == torch::kBFloat16) {
+    return cutlass_scaled_mm_dq_dispatcher<cutlass_2x_gemm<
+        cutlass::arch::Sm75, enable_sm75_to_sm80, int8_t, cutlass::bfloat16_t,
+        TileShape, WarpShape, InstructionShape, 2>>(out, a, b, a_scales,
+                                                    b_scales);
+  } else {
+    TORCH_CHECK(out.dtype() == torch::kFloat16);
+    return cutlass_scaled_mm_dq_dispatcher<cutlass_2x_gemm<
+        cutlass::arch::Sm75, enable_sm75_to_sm80, int8_t, cutlass::half_t,
+        TileShape, WarpShape, InstructionShape, 2>>(out, a, b, a_scales,
+                                                    b_scales);
+  }
+}
+
+void cutlass_scaled_mm_dq_sm80(torch::Tensor& out, torch::Tensor const& a,
+                               torch::Tensor const& b,
+                               torch::Tensor const& a_scales,
+                               torch::Tensor const& b_scales) {
+  TORCH_CHECK(a.dtype() == torch::kInt8);
+  TORCH_CHECK(b.dtype() == torch::kInt8);
+  TORCH_CHECK(a_scales.dtype() == torch::kFloat32);
+  TORCH_CHECK(b_scales.dtype() == torch::kFloat32);
+
+  using TileShape = typename cutlass::gemm::GemmShape<128, 128, 64>;
+  using WarpShape = typename cutlass::gemm::GemmShape<64, 64, 64>;
+  using InstructionShape = typename cutlass::gemm::GemmShape<16, 8, 32>;
+
+  if (out.dtype() == torch::kBFloat16) {
+    return cutlass_scaled_mm_dq_dispatcher<cutlass_2x_gemm<
+        cutlass::arch::Sm80, enable_sm80_to_sm89, int8_t, cutlass::bfloat16_t,
+        TileShape, WarpShape, InstructionShape, 5>>(out, a, b, a_scales,
+                                                    b_scales);
+  } else {
+    TORCH_CHECK(out.dtype() == torch::kFloat16);
+    return cutlass_scaled_mm_dq_dispatcher<cutlass_2x_gemm<
+        cutlass::arch::Sm80, enable_sm80_to_sm89, int8_t, cutlass::half_t,
+        TileShape, WarpShape, InstructionShape, 5>>(out, a, b, a_scales,
+                                                    b_scales);
+  }
+}
+
+void cutlass_scaled_mm_dq_sm89(torch::Tensor& out, torch::Tensor const& a,
+                               torch::Tensor const& b,
+                               torch::Tensor const& a_scales,
+                               torch::Tensor const& b_scales) {
+  using TileShape = typename cutlass::gemm::GemmShape<128, 128, 64>;
+  using WarpShape = typename cutlass::gemm::GemmShape<64, 64, 64>;
+  using InstructionShape = typename cutlass::gemm::GemmShape<16, 8, 32>;
+
+  TORCH_CHECK(a_scales.dtype() == torch::kFloat32);
+  TORCH_CHECK(b_scales.dtype() == torch::kFloat32);
+
+  if (a.dtype() == torch::kInt8) {
+    TORCH_CHECK(b.dtype() == torch::kInt8);
+
+    if (out.dtype() == torch::kBFloat16) {
+      return cutlass_scaled_mm_dq_dispatcher<cutlass_2x_gemm<
+          cutlass::arch::Sm89, enable_sm89_to_sm90, int8_t, cutlass::bfloat16_t,
+          TileShape, WarpShape, InstructionShape, 5>>(out, a, b, a_scales,
+                                                      b_scales);
+    } else {
+      assert(out.dtype() == torch::kFloat16);
+      return cutlass_scaled_mm_dq_dispatcher<cutlass_2x_gemm<
+          cutlass::arch::Sm89, enable_sm89_to_sm90, int8_t, cutlass::half_t,
+          TileShape, WarpShape, InstructionShape, 5>>(out, a, b, a_scales,
+                                                      b_scales);
+    }
+  } else {
+    TORCH_CHECK(a.dtype() == torch::kFloat8_e4m3fn);
+    TORCH_CHECK(b.dtype() == torch::kFloat8_e4m3fn);
+
+    if (out.dtype() == torch::kBFloat16) {
+      return cutlass_scaled_mm_dq_dispatcher<cutlass_2x_gemm<
+          cutlass::arch::Sm89, enable_sm89_to_sm90, cutlass::float_e4m3_t,
+          cutlass::bfloat16_t, TileShape, WarpShape, InstructionShape, 5>>(
+          out, a, b, a_scales, b_scales);
+    } else {
+      TORCH_CHECK(out.dtype() == torch::kFloat16);
+      return cutlass_scaled_mm_dq_dispatcher<cutlass_2x_gemm<
+          cutlass::arch::Sm89, enable_sm89_to_sm90, cutlass::float_e4m3_t,
+          cutlass::half_t, TileShape, WarpShape, InstructionShape, 5>>(
+          out, a, b, a_scales, b_scales);
+    }
+  }
+}

csrc/quantization/cutlass_w8a8/scaled_mm_dq_c3x.cu

@@ -0,0 +1,340 @@
+// clang-format will break include orders
+// clang-format off
+#include <cudaTypedefs.h>
+
+#if defined CUDA_VERSION && CUDA_VERSION >= 12000
+
+#include <torch/all.h>
+
+#include <ATen/cuda/CUDAContext.h>
+
+#include <iostream>
+#include <sstream>
+#include <vector>
+
+#include "cutlass/cutlass.h"
+
+#include "cute/tensor.hpp"
+#include "cute/atom/mma_atom.hpp"
+#include "cutlass/numeric_types.h"
+
+#include "cutlass/util/device_memory.h"
+
+#include "cutlass/gemm/device/gemm_universal_adapter.h"
+#include "cutlass/gemm/kernel/gemm_universal.hpp"
+#include "cutlass/epilogue/collective/collective_builder.hpp"
+#include "cutlass/gemm/collective/collective_builder.hpp"
+
+#include "broadcast_load_epilogue_c3x.hpp"
+#include "common.hpp"
+// clang-format on
+
+using namespace cute;
+
+/*
+   This defines a quantized GEMM operation with dequantized output, similar to
+   torch._scaled_mm. It is defined using the CUTLASS 3.x API, and is used for
+   NVIDIA GPUs with sm90a (Hopper) or later.
+
+   A and B may be both either int8 or fp8_e4m3. A can be quantized per-tensor or
+   per-row. B can be quantized per-tensor or per-column.
+   Any combination of per-tensor and per-row or column is supported.
+   A and B must have symmetric quantization (zero point == 0).
+
+   So the GEMM operation is D = (a_scales * A) (b_scales * B), where the
+   scales are applied elementwise with numpy-style broadcasting.
+
+   ScaleA and ScaleB define the epilogue functions that apply the scales for
+   the A and B operands respectively. These scales may be either per-tensor or
+   per row or column.
+*/
+
+namespace {
+
+uint32_t next_pow_2(uint32_t const num) {
+  if (num <= 1) return num;
+  return 1 << (CHAR_BIT * sizeof(num) - __builtin_clz(num - 1));
+}
+
+// A wrapper for the GEMM kernel that is used to guard against compilation on
+// architectures that will never use the kernel. The purpose of this is to
+// reduce the size of the compiled binary.
+// __CUDA_ARCH__ is not defined in host code, so this lets us smuggle the ifdef
+// into code that will be executed on the device where it is defined.
+template <typename Kernel>
+struct enable_sm90_or_later : Kernel {
+  template <typename... Args>
+  CUTLASS_DEVICE void operator()(Args&&... args) {
+  #if defined __CUDA_ARCH__ && __CUDA_ARCH__ >= 900
+    Kernel::operator()(std::forward<Args>(args)...);
+  #endif
+  }
+};
+
+template <typename ElementAB_, typename ElementD_, typename TileShape,
+          typename ClusterShape, typename KernelSchedule,
+          typename EpilogueSchedule>
+struct cutlass_3x_gemm {
+  using ElementAB = ElementAB_;
+  using ElementD = ElementD_;
+  using ElementAcc =
+      typename std::conditional<std::is_same_v<ElementAB, int8_t>, int32_t,
+                                float>::type;
+
+  using EpilogueDescriptor =
+      cutlass::epilogue::collective::detail::EpilogueDescriptor<
+          TileShape, cutlass::epilogue::collective::EpilogueTileAuto, ElementD,
+          ElementD, EpilogueSchedule>;
+
+  using Accum = cutlass::epilogue::fusion::Sm90AccFetch;
+
+  using ScaleA = cutlass::epilogue::fusion::Sm90ColOrScalarBroadcast<
+      0 /*Stages*/, typename EpilogueDescriptor::TileShape, float,
+      Stride<Int<1>, Int<0>, Int<0>>>;
+
+  using ScaleBDescriptor =
+      cutlass::epilogue::collective::detail::RowBroadcastDescriptor<
+          EpilogueDescriptor, float>;
+
+  using ScaleB = cutlass::epilogue::fusion::Sm90RowOrScalarBroadcast<
+      ScaleBDescriptor::Stages, typename EpilogueDescriptor::TileShape,
+      typename ScaleBDescriptor::Element, Stride<Int<0>, Int<1>, Int<0>>>;
+
+  using Compute0 = cutlass::epilogue::fusion::Sm90Compute<
+      cutlass::multiplies, float, float,
+      cutlass::FloatRoundStyle::round_to_nearest>;
+
+  using EVTCompute0 =
+      cutlass::epilogue::fusion::Sm90EVT<Compute0, ScaleB, Accum>;
+
+  using Compute1 = cutlass::epilogue::fusion::Sm90Compute<
+      cutlass::multiplies, ElementD, float,
+      cutlass::FloatRoundStyle::round_to_nearest>;
+
+  using EVTCompute1 =
+      cutlass::epilogue::fusion::Sm90EVT<Compute1, ScaleA, EVTCompute0>;
+
+  using StrideD = Stride<int64_t, Int<1>, Int<0>>;
+  using ElementC = void;
+  using StrideC = StrideD;
+
+  using CollectiveEpilogue =
+      typename cutlass::epilogue::collective::CollectiveBuilder<
+          cutlass::arch::Sm90, cutlass::arch::OpClassTensorOp, TileShape,
+          ClusterShape, cutlass::epilogue::collective::EpilogueTileAuto,
+          ElementAcc, float, ElementC, StrideC, 4, ElementD, StrideD, 4,
+          EpilogueSchedule, EVTCompute1>::CollectiveOp;
+
+  static constexpr size_t CEStorageSize =
+      sizeof(typename CollectiveEpilogue::SharedStorage);
+  using Stages = typename cutlass::gemm::collective::StageCountAutoCarveout<
+      static_cast<int>(CEStorageSize)>;
+
+  // clang-format off
+  using CollectiveMainloop =
+      typename cutlass::gemm::collective::CollectiveBuilder<
+          cutlass::arch::Sm90, cutlass::arch::OpClassTensorOp, 
+          ElementAB, cutlass::layout::RowMajor, 16, 
+          ElementAB, cutlass::layout::ColumnMajor, 16, 
+          ElementAcc, TileShape, ClusterShape,
+          Stages,
+          KernelSchedule>::CollectiveOp;
+  // clang-format on
+
+  using KernelType = enable_sm90_or_later<cutlass::gemm::kernel::GemmUniversal<
+      cute::Shape<int, int, int, int>, CollectiveMainloop, CollectiveEpilogue,
+      cutlass::gemm::PersistentScheduler>>;
+
+  struct GemmKernel : public KernelType {};
+};
+
+template <typename Gemm>
+void cutlass_scaled_mm_dq_dispatcher(torch::Tensor& out, torch::Tensor const& a,
+                                     torch::Tensor const& b,
+                                     torch::Tensor const& a_scales,
+                                     torch::Tensor const& b_scales) {
+  using ElementAB = typename Gemm::ElementAB;
+  using ElementD = typename Gemm::ElementD;
+
+  int32_t m = a.size(0);
+  int32_t n = b.size(1);
+  int32_t k = a.size(1);
+
+  int64_t lda = a.stride(0);
+  int64_t ldb = b.stride(1);
+  int64_t ldc = out.stride(0);
+
+  using StrideA = Stride<int64_t, Int<1>, Int<0>>;
+  using StrideB = Stride<int64_t, Int<1>, Int<0>>;
+  using StrideC = typename Gemm::StrideC;
+
+  StrideA a_stride{lda, Int<1>{}, Int<0>{}};
+  StrideB b_stride{ldb, Int<1>{}, Int<0>{}};
+  StrideC c_stride{ldc, Int<1>{}, Int<0>{}};
+
+  using GemmKernel = typename Gemm::GemmKernel;
+  typename GemmKernel::ProblemShape prob_shape{m, n, k, 1};
+
+  auto a_ptr = static_cast<ElementAB*>(a.data_ptr());
+  auto b_ptr = static_cast<ElementAB*>(b.data_ptr());
+  typename GemmKernel::MainloopArguments mainloop_args{a_ptr, a_stride, b_ptr,
+                                                       b_stride};
+
+  auto c_ptr = static_cast<ElementD*>(out.data_ptr());
+  typename GemmKernel::EpilogueArguments epilogue_args{
+      {}, c_ptr, c_stride, c_ptr, c_stride};
+
+  typename GemmKernel::Arguments args{cutlass::gemm::GemmUniversalMode::kGemm,
+                                      prob_shape, mainloop_args, epilogue_args};
+
+  using ScaleA_Args = typename Gemm::ScaleA::Arguments;
+  using ScaleB_Args = typename Gemm::ScaleB::Arguments;
+
+  ScaleA_Args a_args{a_scales.data_ptr<float>(), a_scales.numel() != 1, {}};
+  ScaleB_Args b_args{b_scales.data_ptr<float>(), b_scales.numel() != 1, {}};
+
+  args.epilogue.thread = {a_args, {b_args}};
+
+  // Launch the CUTLASS GEMM kernel.
+  using GemmOp = cutlass::gemm::device::GemmUniversalAdapter<GemmKernel>;
+  GemmOp gemm_op;
+  CUTLASS_CHECK(gemm_op.can_implement(args));
+
+  size_t workspace_size = gemm_op.get_workspace_size(args);
+  cutlass::device_memory::allocation<uint8_t> workspace(workspace_size);
+
+  auto stream = at::cuda::getCurrentCUDAStream(a.get_device());
+
+  cutlass::Status status = gemm_op.run(args, workspace.get(), stream);
+  CUTLASS_CHECK(status);
+}
+
+template <typename InType, typename OutType, int32_t M>
+struct sm90_fp8_config {
+  static_assert(std::is_same<InType, cutlass::float_e4m3_t>());
+  using KernelSchedule =
+      cutlass::gemm::KernelTmaWarpSpecializedPingpongFP8FastAccum;
+  using EpilogueSchedule = typename cutlass::epilogue::TmaWarpSpecialized;
+  using TileShape = Shape<_128, _128, _128>;
+  using ClusterShape = Shape<_2, _1, _1>;
+
+  using Cutlass3xGemm =
+      cutlass_3x_gemm<InType, OutType, TileShape, ClusterShape, KernelSchedule,
+                      EpilogueSchedule>;
+};
+
+template <typename InType, typename OutType>
+struct sm90_fp8_config<InType, OutType, 128> {
+  static_assert(std::is_same<InType, cutlass::float_e4m3_t>());
+  using KernelSchedule =
+      cutlass::gemm::KernelTmaWarpSpecializedPingpongFP8FastAccum;
+  using EpilogueSchedule = typename cutlass::epilogue::TmaWarpSpecialized;
+  using TileShape = Shape<_64, _128, _128>;
+  using ClusterShape = Shape<_2, _1, _1>;
+
+  using Cutlass3xGemm =
+      cutlass_3x_gemm<InType, OutType, TileShape, ClusterShape, KernelSchedule,
+                      EpilogueSchedule>;
+};
+
+template <typename InType, typename OutType>
+struct sm90_fp8_config<InType, OutType, 64> {
+  static_assert(std::is_same<InType, cutlass::float_e4m3_t>());
+  using KernelSchedule =
+      cutlass::gemm::KernelTmaWarpSpecializedPingpongFP8FastAccum;
+  using EpilogueSchedule = typename cutlass::epilogue::TmaWarpSpecialized;
+  using TileShape = Shape<_64, _64, _128>;
+  using ClusterShape = Shape<_1, _8, _1>;
+
+  using Cutlass3xGemm =
+      cutlass_3x_gemm<InType, OutType, TileShape, ClusterShape, KernelSchedule,
+                      EpilogueSchedule>;
+};
+
+}  // namespace
+
+template <typename InType, typename OutType>
+void cutlass_scaled_mm_dq_sm90_fp8_dispatch(torch::Tensor& out,
+                                            torch::Tensor const& a,
+                                            torch::Tensor const& b,
+                                            torch::Tensor const& a_scales,
+                                            torch::Tensor const& b_scales) {
+  static_assert(std::is_same<InType, cutlass::float_e4m3_t>());
+  TORCH_CHECK(a.dtype() == torch::kFloat8_e4m3fn);
+  TORCH_CHECK(b.dtype() == torch::kFloat8_e4m3fn);
+  TORCH_CHECK(a_scales.dtype() == torch::kFloat32);
+  TORCH_CHECK(b_scales.dtype() == torch::kFloat32);
+
+  using Cutlass3xGemmDefault =
+      typename sm90_fp8_config<InType, OutType, 0>::Cutlass3xGemm;
+  using Cutlass3xGemmM64 =
+      typename sm90_fp8_config<InType, OutType, 64>::Cutlass3xGemm;
+  using Cutlass3xGemmM128 =
+      typename sm90_fp8_config<InType, OutType, 128>::Cutlass3xGemm;
+
+  uint32_t const m = a.size(0);
+  uint32_t const mp2 =
+      std::max(static_cast<uint32_t>(64), next_pow_2(m));  // next power of 2
+
+  if (mp2 <= 64) {
+    // m in [1, 64]
+    return cutlass_scaled_mm_dq_dispatcher<Cutlass3xGemmM64>(
+        out, a, b, a_scales, b_scales);
+  } else if (mp2 <= 128) {
+    // m in (64, 128]
+    return cutlass_scaled_mm_dq_dispatcher<Cutlass3xGemmM128>(
+        out, a, b, a_scales, b_scales);
+  } else {
+    // m in (128, inf)
+    return cutlass_scaled_mm_dq_dispatcher<Cutlass3xGemmDefault>(
+        out, a, b, a_scales, b_scales);
+  }
+}
+
+void cutlass_scaled_mm_dq_sm90(torch::Tensor& out, torch::Tensor const& a,
+                               torch::Tensor const& b,
+                               torch::Tensor const& a_scales,
+                               torch::Tensor const& b_scales) {
+  TORCH_CHECK(a_scales.dtype() == torch::kFloat32);
+  TORCH_CHECK(b_scales.dtype() == torch::kFloat32);
+
+  if (a.dtype() == torch::kInt8) {
+    TORCH_CHECK(b.dtype() == torch::kInt8);
+
+    using TileShape = Shape<_128, _128, _128>;
+    using ClusterShape = Shape<_1, _2, _1>;
+    using KernelSchedule =
+        typename cutlass::gemm::KernelTmaWarpSpecializedPingpong;
+    using EpilogueSchedule = typename cutlass::epilogue::TmaWarpSpecialized;
+
+    if (out.dtype() == torch::kBFloat16) {
+      return cutlass_scaled_mm_dq_dispatcher<
+          cutlass_3x_gemm<int8_t, cutlass::bfloat16_t, TileShape, ClusterShape,
+                          KernelSchedule, EpilogueSchedule>>(
+          out, a, b, a_scales, b_scales);
+    } else {
+      TORCH_CHECK(out.dtype() == torch::kFloat16);
+
+      return cutlass_scaled_mm_dq_dispatcher<
+          cutlass_3x_gemm<int8_t, cutlass::half_t, TileShape, ClusterShape,
+                          KernelSchedule, EpilogueSchedule>>(
+          out, a, b, a_scales, b_scales);
+    }
+  } else {
+    TORCH_CHECK(a.dtype() == torch::kFloat8_e4m3fn);
+    TORCH_CHECK(b.dtype() == torch::kFloat8_e4m3fn);
+
+    if (out.dtype() == torch::kBFloat16) {
+      return cutlass_scaled_mm_dq_sm90_fp8_dispatch<cutlass::float_e4m3_t,
+                                                    cutlass::bfloat16_t>(
+          out, a, b, a_scales, b_scales);
+    } else {
+      TORCH_CHECK(out.dtype() == torch::kFloat16);
+      return cutlass_scaled_mm_dq_sm90_fp8_dispatch<cutlass::float_e4m3_t,
+                                                    cutlass::half_t>(
+          out, a, b, a_scales, b_scales);
+    }
+  }
+}
+
+#endif

csrc/quantization/cutlass_w8a8/scaled_mm_dq_entry.cu

@@ -0,0 +1,75 @@
+#include <cudaTypedefs.h>
+
+#include <c10/cuda/CUDAGuard.h>
+#include <torch/all.h>
+
+void cutlass_scaled_mm_dq_sm75(torch::Tensor& c, torch::Tensor const& a,
+                               torch::Tensor const& b,
+                               torch::Tensor const& a_scales,
+                               torch::Tensor const& b_scales);
+
+void cutlass_scaled_mm_dq_sm80(torch::Tensor& c, torch::Tensor const& a,
+                               torch::Tensor const& b,
+                               torch::Tensor const& a_scales,
+                               torch::Tensor const& b_scales);
+
+void cutlass_scaled_mm_dq_sm89(torch::Tensor& c, torch::Tensor const& a,
+                               torch::Tensor const& b,
+                               torch::Tensor const& a_scales,
+                               torch::Tensor const& b_scales);
+
+#if defined CUDA_VERSION && CUDA_VERSION >= 12000
+void cutlass_scaled_mm_dq_sm90(torch::Tensor& c, torch::Tensor const& a,
+                               torch::Tensor const& b,
+                               torch::Tensor const& a_scales,
+                               torch::Tensor const& b_scales);
+#endif
+
+void cutlass_scaled_mm_dq(torch::Tensor& c, torch::Tensor const& a,
+                          torch::Tensor const& b, torch::Tensor const& a_scales,
+                          torch::Tensor const& b_scales) {
+  int32_t major_capability;
+  int32_t minor_capability;
+  cudaDeviceGetAttribute(&major_capability, cudaDevAttrComputeCapabilityMajor,
+                         0);
+  cudaDeviceGetAttribute(&minor_capability, cudaDevAttrComputeCapabilityMinor,
+                         0);
+  int32_t version_num = major_capability * 10 + minor_capability;
+
+  // Checks for conformality
+  TORCH_CHECK(a.dim() == 2 && b.dim() == 2 && c.dim() == 2);
+  TORCH_CHECK(c.size(0) == a.size(0) && a.size(1) == b.size(0) &&
+              b.size(1) == c.size(1));
+  TORCH_CHECK(a_scales.numel() == 1 || a_scales.numel() == a.size(0));
+  TORCH_CHECK(b_scales.numel() == 1 || b_scales.numel() == b.size(1));
+
+  // Check for strides and alignment
+  TORCH_CHECK(a.stride(1) == 1 && c.stride(1) == 1);  // Row-major
+  TORCH_CHECK(b.stride(0) == 1);                      // Column-major
+  TORCH_CHECK(c.stride(0) % 16 == 0 &&
+              b.stride(1) % 16 == 0);  // 16 Byte Alignment
+  TORCH_CHECK(a_scales.is_contiguous() && b_scales.is_contiguous());
+
+  at::cuda::OptionalCUDAGuard const device_guard(device_of(a));
+
+  if (version_num >= 90) {
+    // Hopper
+
+    // Guard against compilation issues for sm90 kernels
+#if defined CUDA_VERSION && CUDA_VERSION >= 12000
+    cutlass_scaled_mm_dq_sm90(c, a, b, a_scales, b_scales);
+#else
+    cutlass_scaled_mm_dq_sm80(c, a, b, a_scales, b_scales);
+#endif
+  } else if (version_num == 89) {
+    // Ada Lovelace
+    cutlass_scaled_mm_dq_sm89(c, a, b, a_scales, b_scales);
+  } else if (version_num >= 80) {
+    // Ampere
+    cutlass_scaled_mm_dq_sm80(c, a, b, a_scales, b_scales);
+  } else {
+    // Turing
+    TORCH_CHECK(version_num >= 75);
+    cutlass_scaled_mm_dq_sm75(c, a, b, a_scales, b_scales);
+  }
+}