[V1] Minor V1 async engine test refactor (#15075)

Signed-off-by: andoorve <murali.andoorveedu@mail.utoronto.ca>
Co-authored-by: andoorve <murali.andoorveedu@mail.utoronto.ca>

.buildkite/run-tpu-test.sh

@@ -1,25 +0,0 @@
-#!/bin/bash
-
-set -e
-
-# Build the docker image.
-docker build -f Dockerfile.tpu -t vllm-tpu .
-
-# Set up cleanup.
-remove_docker_container() { docker rm -f tpu-test || true; }
-trap remove_docker_container EXIT
-# Remove the container that might not be cleaned up in the previous run.
-remove_docker_container
-
-# For HF_TOKEN.
-source /etc/environment
-# Run a simple end-to-end example.
-docker run --privileged --net host --shm-size=16G -it \
-    -e "HF_TOKEN=$HF_TOKEN" --name tpu-test \
-    vllm-tpu /bin/bash -c "python3 -m pip install git+https://github.com/thuml/depyf.git \
-    && python3 -m pip install pytest \
-    && python3 -m pip install lm_eval[api]==0.4.4 \
-    && pytest -v -s /workspace/vllm/tests/tpu/test_custom_dispatcher.py \
-    && python3 /workspace/vllm/tests/tpu/test_compilation.py \
-    && python3 /workspace/vllm/tests/tpu/test_quantization_accuracy.py \
-    && python3 /workspace/vllm/examples/offline_inference/tpu.py"

.buildkite/run-tpu-v1-test.sh

@@ -15,13 +15,22 @@ remove_docker_container
 source /etc/environment
 # Run a simple end-to-end example.
 docker run --privileged --net host --shm-size=16G -it \
-    -e "HF_TOKEN=$HF_TOKEN" -e "VLLM_USE_V1=1" --name tpu-test \
+    -e "HF_TOKEN=$HF_TOKEN" --name tpu-test \
     vllm-tpu /bin/bash -c "python3 -m pip install git+https://github.com/thuml/depyf.git \
     && python3 -m pip install pytest \
     && python3 -m pip install lm_eval[api]==0.4.4 \
-    && pytest -v -s /workspace/vllm/tests/tpu/test_custom_dispatcher.py \
+    && echo TEST_1 \
-    && pytest -v -s /workspace/vllm/tests/v1/tpu/test_basic.py \
+    && VLLM_USE_V1=1 python3 /workspace/vllm/tests/tpu/test_compilation.py \
-    && pytest -v -s /workspace/vllm/tests/entrypoints/llm/test_accuracy.py::test_lm_eval_accuracy_v1_engine \
+    && echo TEST_2 \
-    && python3 /workspace/vllm/tests/tpu/test_compilation.py \
+    && VLLM_USE_V1=1 pytest -v -s /workspace/vllm/tests/v1/tpu/test_basic.py \
-    && python3 /workspace/vllm/tests/tpu/test_quantization_accuracy.py \
+    && echo TEST_3 \
-    && python3 /workspace/vllm/examples/offline_inference/tpu.py"
+    && VLLM_USE_V1=1 pytest -v -s /workspace/vllm/tests/entrypoints/llm/test_accuracy.py::test_lm_eval_accuracy_v1_engine \
+    && echo TEST_4 \
+    && VLLM_USE_V1=1 pytest -s -v /workspace/vllm/tests/tpu/test_quantization_accuracy.py \
+    && echo TEST_5 \
+    && VLLM_USE_V1=1 python3 /workspace/vllm/examples/offline_inference/tpu.py" \
+    
+
+# TODO: This test fails because it uses RANDOM_SEED sampling
+# && VLLM_USE_V1=1 pytest -v -s /workspace/vllm/tests/tpu/test_custom_dispatcher.py \
+

.buildkite/run-xpu-test.sh

@@ -12,10 +12,11 @@ docker build -t ${image_name} -f Dockerfile.xpu .
 
 # Setup cleanup
 remove_docker_container() { 
-  docker rm -f "${container_name}" || docker image rm -f "${image_name}" || true;
+  docker rm -f "${container_name}" || true; 
+  docker image rm -f "${image_name}" || true;
+  docker system prune -f || true;
 }
 trap remove_docker_container EXIT
-remove_docker_container
 
 # Run the image and test offline inference/tensor parallel
 docker run \
@@ -25,6 +26,6 @@ docker run \
     --name "${container_name}" \
     "${image_name}" \
     sh -c '
-    python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m
+    VLLM_USE_V1=0 python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m
-    python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m -tp 2
+    VLLM_USE_V1=0 python3 examples/offline_inference/basic/generate.py --model facebook/opt-125m -tp 2
 '

Dockerfile.xpu

@@ -1,11 +1,7 @@
-FROM intel/deep-learning-essentials:2025.0.1-0-devel-ubuntu22.04 AS vllm-base
+# oneapi 2025.0.2 docker base image use rolling 2448 package. https://dgpu-docs.intel.com/releases/packages.html?release=Rolling+2448.13&os=Ubuntu+22.04, and we don't need install driver manually.
+FROM intel/deep-learning-essentials:2025.0.2-0-devel-ubuntu22.04 AS vllm-base
 
-RUN wget -O- https://apt.repos.intel.com/intel-gpg-keys/GPG-PUB-KEY-INTEL-SW-PRODUCTS.PUB | gpg --dearmor | tee /usr/share/keyrings/intel-oneapi-archive-keyring.gpg > /dev/null && \
+RUN rm /etc/apt/sources.list.d/intel-graphics.list
-    echo "deb [signed-by=/usr/share/keyrings/intel-oneapi-archive-keyring.gpg] https://apt.repos.intel.com/oneapi all main " | tee /etc/apt/sources.list.d/oneAPI.list && \
-    chmod 644 /usr/share/keyrings/intel-oneapi-archive-keyring.gpg && \
-    wget -O- https://repositories.intel.com/graphics/intel-graphics.key | gpg --dearmor | tee /usr/share/keyrings/intel-graphics.gpg > /dev/null && \
-    echo "deb [arch=amd64,i386 signed-by=/usr/share/keyrings/intel-graphics.gpg] https://repositories.intel.com/graphics/ubuntu jammy arc" | tee /etc/apt/sources.list.d/intel.gpu.jammy.list && \
-    chmod 644 /usr/share/keyrings/intel-graphics.gpg
 
 RUN apt-get update -y && \
     apt-get install -y --no-install-recommends --fix-missing \
@@ -21,8 +17,6 @@ RUN apt-get update -y && \
     python3 \
     python3-dev \
     python3-pip \
-    libze-intel-gpu-dev \
-    libze-intel-gpu1 \
     wget
 
 WORKDIR /workspace/vllm

README.md

@@ -15,7 +15,7 @@ Easy, fast, and cheap LLM serving for everyone
 
 *Latest News* 🔥
 
-- [2025/03] We hosted [the first vLLM China Meetup](https://mp.weixin.qq.com/s/n77GibL2corAtQHtVEAzfg)! Please find the meetup slides from vLLM team [here](https://docs.google.com/presentation/d/1REHvfQMKGnvz6p3Fd23HhSO4c8j5WPGZV0bKYLwnHyQ/edit#slide=id.g33fb1ff286e_0_29).
+- [2025/03] We hosted [the first vLLM China Meetup](https://mp.weixin.qq.com/s/n77GibL2corAtQHtVEAzfg)! Please find the meetup slides from vLLM team [here](https://docs.google.com/presentation/d/1REHvfQMKGnvz6p3Fd23HhSO4c8j5WPGZV0bKYLwnHyQ/edit?usp=sharing).
 - [2025/03] We hosted [the East Coast vLLM Meetup](https://lu.ma/7mu4k4xx)! Please find the meetup slides [here](https://docs.google.com/presentation/d/1NHiv8EUFF1NLd3fEYODm56nDmL26lEeXCaDgyDlTsRs/edit#slide=id.g31441846c39_0_0).
 - [2025/02] We hosted [the ninth vLLM meetup](https://lu.ma/h7g3kuj9) with Meta! Please find the meetup slides from vLLM team [here](https://docs.google.com/presentation/d/1jzC_PZVXrVNSFVCW-V4cFXb6pn7zZ2CyP_Flwo05aqg/edit?usp=sharing) and AMD [here](https://drive.google.com/file/d/1Zk5qEJIkTmlQ2eQcXQZlljAx3m9s7nwn/view?usp=sharing). The slides from Meta will not be posted.
 - [2025/01] We are excited to announce the alpha release of vLLM V1: A major architectural upgrade with 1.7x speedup! Clean code, optimized execution loop, zero-overhead prefix caching, enhanced multimodal support, and more. Please check out our blog post [here](https://blog.vllm.ai/2025/01/27/v1-alpha-release.html).

benchmarks/benchmark_serving_structured_output.py

@@ -732,8 +732,11 @@ def main(args: argparse.Namespace):
         api_url = f"http://{args.host}:{args.port}{args.endpoint}"
         base_url = f"http://{args.host}:{args.port}"
 
-    tokenizer = get_tokenizer(tokenizer_id,
+    tokenizer = get_tokenizer(
-                              trust_remote_code=args.trust_remote_code)
+        tokenizer_id,
+        trust_remote_code=args.trust_remote_code,
+        tokenizer_mode=args.tokenizer_mode,
+    )
 
     if args.dataset == 'grammar':
         args.structure_type = 'guided_grammar'
@@ -876,6 +879,13 @@ if __name__ == "__main__":
         help=
         "Name or path of the tokenizer, if not using the default tokenizer.",  # noqa: E501
     )
+    parser.add_argument(
+        "--tokenizer-mode",
+        type=str,
+        default="auto",
+        help=
+        "Name or path of the tokenizer, if not using the default tokenizer.",  # noqa: E501
+    )
     parser.add_argument(
         "--num-prompts",
         type=int,

benchmarks/run_structured_output_benchmark.sh

@@ -54,6 +54,7 @@ for qps in "${QPS_VALUES[@]}"; do
   python "$SCRIPT_DIR/benchmark_serving_structured_output.py" $COMMON_PARAMS \
     --request-rate $qps \
     --result-filename "$FILENAME" \
+    --tokenizer-mode ${TOKENIZER_MODE:-"auto"} \
     --port ${PORT:-8000}
 
   echo "Completed benchmark with QPS: $qps"

csrc/cache_kernels.cu

@@ -350,8 +350,8 @@ __global__ void concat_and_cache_mla_kernel(
 
 }  // namespace vllm
 
-// KV_T is the stored data type of kv-cache.
+// KV_T is the data type of key and value tensors.
-// CACHE_T is the data type of key and value tensors.
+// CACHE_T is the stored data type of kv-cache.
 // 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>             \
@@ -393,8 +393,8 @@ void reshape_and_cache(
                              CALL_RESHAPE_AND_CACHE)
 }
 
-// KV_T is the stored data type of kv-cache.
+// KV_T is the data type of key and value tensors.
-// CACHE_T is the data type of key and value tensors.
+// CACHE_T is the stored data type of kv-cache.
 // KV_DTYPE is the real data type of kv-cache.
 #define CALL_RESHAPE_AND_CACHE_FLASH(KV_T, CACHE_T, KV_DTYPE)         \
   vllm::reshape_and_cache_flash_kernel<KV_T, CACHE_T, KV_DTYPE>       \
@@ -446,8 +446,8 @@ void reshape_and_cache_flash(
                              CALL_RESHAPE_AND_CACHE_FLASH);
 }
 
-// KV_T is the stored data type of kv-cache.
+// KV_T is the data type of key and value tensors.
-// CACHE_T is the data type of key and value tensors.
+// CACHE_T is the stored data type of kv-cache.
 // KV_DTYPE is the real data type of kv-cache.
 #define CALL_CONCAT_AND_CACHE_MLA(KV_T, CACHE_T, KV_DTYPE)              \
   vllm::concat_and_cache_mla_kernel<KV_T, CACHE_T, KV_DTYPE>            \

docs/source/deployment/docker.md

@@ -34,11 +34,11 @@ If you need to use those dependencies (having accepted the license terms),
 create a custom Dockerfile on top of the base image with an extra layer that installs them:
 
 ```Dockerfile
-FROM vllm/vllm-openai:v0.7.3
+FROM vllm/vllm-openai:v0.8.0
 
 # e.g. install the `audio` and `video` optional dependencies
 # NOTE: Make sure the version of vLLM matches the base image!
-RUN uv pip install --system vllm[audio,video]==0.7.3
+RUN uv pip install vllm[audio,video]==0.8.0
 ```
 
 :::
@@ -52,7 +52,7 @@ with an extra layer that installs their code from source:
 ```Dockerfile
 FROM vllm/vllm-openai:latest
 
-RUN uv pip install --system git+https://github.com/huggingface/transformers.git
+RUN uv pip install git+https://github.com/huggingface/transformers.git
 ```
 
 :::

docs/source/design/v1/prefix_caching.md

@@ -191,7 +191,7 @@ When the head block (least recently used block) of the free queue is cached, we
 
 In this example, we assume the block size is 4 (each block can cache 4 tokens), and we have 10 blocks in the KV-cache manager in total.
 
-**Time 1: The cache is empty and a new request comes in.** We allocate 4 blocks. 3 of them are already full and cached. The fourth block is partially full with 2 of 4 tokens.
+**Time 1: The cache is empty and a new request comes in.** We allocate 4 blocks. 3 of them are already full and cached. The fourth block is partially full with 3 of 4 tokens.
 
 :::{image} /assets/design/v1/prefix_caching/example-time-1.png
 :alt: Example Time 1
@@ -203,7 +203,7 @@ In this example, we assume the block size is 4 (each block can cache 4 tokens),
 :alt: Example Time 3
 :::
 
-**Time 4: Request 1 comes in with the 14 prompt tokens, where the first 11 tokens are the same as request 0.** We can see that only 2 blocks (11 tokens) hit the cache, because the 3rd block only matches 3 of 4 tokens.
+**Time 4: Request 1 comes in with the 14 prompt tokens, where the first 10 tokens are the same as request 0.** We can see that only the first 2 blocks (8 tokens) hit the cache, because the 3rd block only matches 2 of 4 tokens.
 
 :::{image} /assets/design/v1/prefix_caching/example-time-4.png
 :alt: Example Time 4

docs/source/getting_started/v1_user_guide.md

@@ -2,6 +2,8 @@
 
 V1 is now enabled by default for all supported use cases, and we will gradually enable it for every use case we plan to support. Please share any feedback on [GitHub](https://github.com/vllm-project/vllm) or in the [vLLM Slack](https://inviter.co/vllm-slack).
 
+To disable V1, please set the environment variable as: `VLLM_USE_V1=0`, and send us a GitHub issue sharing the reason!
+
 ## Why vLLM V1?
 
 vLLM V0 successfully supported a wide range of models and hardware, but as new features were developed independently, the system grew increasingly complex. This complexity made it harder to integrate new capabilities and introduced technical debt, revealing the need for a more streamlined and unified design.

docs/source/models/supported_models.md

@@ -768,7 +768,7 @@ See [this page](#generative-models) for more information on how to use generativ
   * `google/gemma-3-4b-it`, `google/gemma-3-27b-it`, etc.
   * ✅︎
   * ✅︎
-  *
+  * ⚠️
 - * `GLM4VForCausalLM`<sup>^</sup>
   * GLM-4V
   * T + I
@@ -884,7 +884,7 @@ See [this page](#generative-models) for more information on how to use generativ
 - * `PixtralForConditionalGeneration`
   * Pixtral
   * T + I<sup>+</sup>
-  * `mistralai/Pixtral-12B-2409`, `mistral-community/pixtral-12b`, etc.
+  * `mistralai/Mistral-Small-3.1-24B-Instruct-2503`, `mistral-community/pixtral-12b`, etc.
   *
   * ✅︎
   * ✅︎
@@ -951,13 +951,10 @@ V0 correctly implements the model's attention pattern:
 
 V1 currently uses a simplified attention pattern:
 - Uses causal attention for all tokens, including image tokens
-- Generates reasonable outputs but does not match the original model's attention for text + image inputs
+- Generates reasonable outputs but does not match the original model's attention for text + image inputs, especially when `{"do_pan_and_scan": True}`
 - Will be updated in the future to support the correct behavior
-- Does not support `"do_pan_and_scan": True`
 
 This limitation exists because the model's mixed attention pattern (bidirectional for images, causal otherwise) is not yet supported by vLLM's attention backends.
-
-For these reasons, `Gemma3ForConditionalGeneration` is supported only on V0 at the moment.
 :::
 
 :::{note}

examples/offline_inference/pixtral.py

@@ -6,14 +6,14 @@ import argparse
 from vllm import LLM
 from vllm.sampling_params import SamplingParams
 
-# This script is an offline demo for running Pixtral.
+# This script is an offline demo for running Mistral-Small-3
 #
 # If you want to run a server/client setup, please follow this code:
 #
 # - Server:
 #
 # ```bash
-# vllm serve mistralai/Pixtral-12B-2409 --tokenizer-mode mistral --limit-mm-per-prompt 'image=4' --max-model-len 16384
+# vllm serve mistralai/Mistral-Small-3.1-24B-Instruct-2503 --tokenizer-mode mistral --limit-mm-per-prompt 'image=4' --max-model-len 16384
 # ```
 #
 # - Client:
@@ -23,7 +23,7 @@ from vllm.sampling_params import SamplingParams
 # --header 'Content-Type: application/json' \
 # --header 'Authorization: Bearer token' \
 # --data '{
-#     "model": "mistralai/Pixtral-12B-2409",
+#     "model": "mistralai/Mistral-Small-3.1-24B-Instruct-2503",
 #     "messages": [
 #       {
 #         "role": "user",
@@ -44,7 +44,7 @@ from vllm.sampling_params import SamplingParams
 
 
 def run_simple_demo(args: argparse.Namespace):
-    model_name = "mistralai/Pixtral-12B-2409"
+    model_name = "mistralai/Mistral-Small-3.1-24B-Instruct-2503"
     sampling_params = SamplingParams(max_tokens=8192)
 
     # Lower max_model_len and/or max_num_seqs on low-VRAM GPUs.
@@ -83,7 +83,7 @@ def run_simple_demo(args: argparse.Namespace):
 
 
 def run_advanced_demo(args: argparse.Namespace):
-    model_name = "mistralai/Pixtral-12B-2409"
+    model_name = "mistralai/Mistral-Small-3.1-24B-Instruct-2503"
     max_img_per_msg = 5
     max_tokens_per_img = 4096
 

tests/compile/test_basic_correctness.py

@@ -60,7 +60,7 @@ class TestSetting:
         # embedding model
         TestSetting(
             model="BAAI/bge-multilingual-gemma2",
-            model_args=["--task", "embed"],
+            model_args=["--task", "embed", "--dtype", "bfloat16"],
             pp_size=1,
             tp_size=1,
             attn_backend="FLASH_ATTN",

tests/conftest.py

@@ -14,8 +14,8 @@ import torch.nn as nn
 import torch.nn.functional as F
 from huggingface_hub import snapshot_download
 from PIL import Image
-from transformers import (AutoModelForCausalLM, AutoTokenizer, BatchEncoding,
+from transformers import (AutoConfig, AutoModelForCausalLM, AutoTokenizer,
-                          BatchFeature)
+                          BatchEncoding, BatchFeature)
 from transformers.models.auto.auto_factory import _BaseAutoModelClass
 
 from tests.models.utils import (TokensTextLogprobs,
@@ -23,7 +23,7 @@ from tests.models.utils import (TokensTextLogprobs,
 from vllm import LLM, SamplingParams
 from vllm.assets.image import ImageAsset
 from vllm.assets.video import VideoAsset
-from vllm.config import TaskOption, TokenizerPoolConfig
+from vllm.config import TaskOption, TokenizerPoolConfig, _get_and_verify_dtype
 from vllm.connections import global_http_connection
 from vllm.distributed import (cleanup_dist_env_and_memory,
                               init_distributed_environment,
@@ -34,8 +34,7 @@ from vllm.inputs import (ExplicitEncoderDecoderPrompt, TextPrompt,
 from vllm.logger import init_logger
 from vllm.outputs import RequestOutput
 from vllm.sampling_params import BeamSearchParams
-from vllm.utils import (STR_DTYPE_TO_TORCH_DTYPE, cuda_device_count_stateless,
+from vllm.utils import cuda_device_count_stateless, is_list_of
-                        identity, is_list_of)
 
 logger = init_logger(__name__)
 
@@ -271,14 +270,18 @@ _R = TypeVar("_R")
 
 class HfRunner:
 
-    def wrap_device(self, x: _T, device: Optional[str] = None) -> _T:
+    def get_default_device(self):
         from vllm.platforms import current_platform
+
+        return ("cpu" if current_platform.is_cpu()
+                or current_platform.is_openvino() else "cuda")
+
+    def wrap_device(self, x: _T, device: Optional[str] = None) -> _T:
         if x is None or isinstance(x, (bool, )):
             return x
 
         if device is None:
-            device = "cpu" if current_platform.is_cpu(
+            device = self.device
-            ) or current_platform.is_openvino() else "cuda"
 
         if isinstance(x, dict):
             return {k: self.wrap_device(v, device) for k, v in x.items()}
@@ -291,45 +294,59 @@ class HfRunner:
     def __init__(
         self,
         model_name: str,
-        dtype: str = "half",
+        dtype: str = "auto",
         *,
         model_kwargs: Optional[dict[str, Any]] = None,
         is_sentence_transformer: bool = False,
         is_cross_encoder: bool = False,
         skip_tokenizer_init: bool = False,
         auto_cls: type[_BaseAutoModelClass] = AutoModelForCausalLM,
-        postprocess_inputs: Callable[..., BatchEncoding] = identity,
     ) -> None:
-        torch_dtype = STR_DTYPE_TO_TORCH_DTYPE[dtype]
-
         self.model_name = model_name
 
+        self.config = AutoConfig.from_pretrained(
+            model_name,
+            trust_remote_code=True,
+        )
+        self.device = self.get_default_device()
+        self.dtype = torch_dtype = _get_and_verify_dtype(self.config, dtype)
+
+        model_kwargs = model_kwargs if model_kwargs is not None else {}
+        model_kwargs.setdefault("torch_dtype", torch_dtype)
+
         if is_sentence_transformer:
             # Lazy init required for AMD CI
             from sentence_transformers import SentenceTransformer
-            self.model = self.wrap_device(
+
-                SentenceTransformer(
+            self.model = SentenceTransformer(
                 model_name,
-                    device="cpu",
+                device=self.device,
+                model_kwargs=model_kwargs,
                 trust_remote_code=True,
-                ).to(dtype=torch_dtype))
+            )
         elif is_cross_encoder:
             # Lazy init required for AMD CI
             from sentence_transformers import CrossEncoder
-            self.model = CrossEncoder(model_name,
+
-                                      device="cpu",
+            self.model = CrossEncoder(
-                                      trust_remote_code=True)
+                model_name,
-            self.model.model = self.wrap_device(self.model.model)\
+                device=self.device,
-                .to(dtype=torch_dtype)
+                automodel_args=model_kwargs,
-        else:
+                trust_remote_code=True,
-            model_kwargs = model_kwargs if model_kwargs is not None else {}
+            )
-            self.model = self.wrap_device(
+        else:
-                auto_cls.from_pretrained(
+            model = auto_cls.from_pretrained(
                 model_name,
-                    torch_dtype=torch_dtype,
                 trust_remote_code=True,
                 **model_kwargs,
-                ))
+            )
+
+            if (getattr(model, "quantization_method", None) != "bitsandbytes"
+                    and len({p.device
+                             for p in model.parameters()}) < 2):
+                model = model.to(self.device)
+
+            self.model = model
 
         if not skip_tokenizer_init:
             self.tokenizer = AutoTokenizer.from_pretrained(
@@ -349,16 +366,13 @@ class HfRunner:
         if skip_tokenizer_init:
             self.tokenizer = self.processor.tokenizer
 
-        self.dtype = dtype
-        self.postprocess_inputs = postprocess_inputs
-
     def get_inputs(
         self,
         prompts: list[str],
         images: Optional[PromptImageInput] = None,
         videos: Optional[PromptVideoInput] = None,
         audios: Optional[PromptAudioInput] = None,
-    ) -> list[BatchEncoding]:
+    ) -> list[Union[BatchFeature, BatchEncoding]]:
         if images is not None:
             assert len(prompts) == len(images)
 
@@ -368,7 +382,7 @@ class HfRunner:
         if audios is not None:
             assert len(prompts) == len(audios)
 
-        all_inputs: list[BatchEncoding] = []
+        all_inputs: list[Union[BatchFeature, BatchEncoding]] = []
         for i, prompt in enumerate(prompts):
             processor_kwargs: dict[str, Any] = {
                 "text": prompt,
@@ -384,7 +398,8 @@ class HfRunner:
                 processor_kwargs["sampling_rate"] = sr
 
             inputs = self.processor(**processor_kwargs)
-            inputs = self.postprocess_inputs(inputs, dtype=self.dtype)
+            if isinstance(inputs, BatchFeature):
+                inputs = inputs.to(dtype=self.dtype)
 
             all_inputs.append(inputs)
 
@@ -417,7 +432,7 @@ class HfRunner:
         outputs: list[tuple[list[list[int]], list[str]]] = []
         for inputs in all_inputs:
             output_ids = self.model.generate(
-                **self.wrap_device(inputs, device=self.model.device.type),
+                **self.wrap_device(inputs),
                 use_cache=True,
                 **kwargs,
             )
@@ -488,7 +503,7 @@ class HfRunner:
         all_logprobs: list[list[torch.Tensor]] = []
         for inputs in all_inputs:
             output = self.model.generate(
-                **self.wrap_device(inputs, device=self.model.device.type),
+                **self.wrap_device(inputs),
                 use_cache=True,
                 do_sample=False,
                 max_new_tokens=max_tokens,
@@ -569,7 +584,7 @@ class HfRunner:
 
         for inputs in all_inputs:
             output = self.model.generate(
-                **self.wrap_device(inputs, device=self.model.device.type),
+                **self.wrap_device(inputs),
                 use_cache=True,
                 do_sample=False,
                 max_new_tokens=max_tokens,
@@ -620,19 +635,15 @@ class HfRunner:
             if images is not None and images[i] is not None:
                 processor_kwargs["images"] = images[i]
 
-            encoder_inputs = self.wrap_device(
+            encoder_inputs = self.processor(**processor_kwargs)
-                self.processor(**processor_kwargs),
+            encoder_inputs = self.wrap_device(encoder_inputs)
-                device=self.model.device.type,
-            )
 
             if decoder_prompt is None:
                 decoder_input_ids = None
             else:
-                decoder_input_ids = self.wrap_device(
+                decoder_inputs = self.tokenizer(decoder_prompt,
-                    self.tokenizer(decoder_prompt,
+                                                return_tensors="pt")
-                                   return_tensors="pt").input_ids,
+                decoder_input_ids = self.wrap_device(decoder_inputs.input_ids)
-                    device=self.model.device.type,
-                )
 
             output = self.model.generate(
                 decoder_input_ids=decoder_input_ids,
@@ -684,6 +695,7 @@ class VllmRunner:
     """
     The default value of some arguments have been modified from
     :class:`~vllm.LLM` as follows:
+
     - `trust_remote_code`: Set to `True` instead of `False` for convenience.
     - `seed`: Set to `0` instead of `None` for test reproducibility.
     - `max_model_len`: Set to `1024` instead of `None` to reduce memory usage.
@@ -701,10 +713,8 @@ class VllmRunner:
         tokenizer_mode: str = "auto",
         trust_remote_code: bool = True,
         seed: Optional[int] = 0,
-        # Use smaller max model length, otherwise bigger model cannot run due
-        # to kv cache size limit.
         max_model_len: int = 1024,
-        dtype: str = "half",
+        dtype: str = "auto",
         disable_log_stats: bool = True,
         tensor_parallel_size: int = 1,
         block_size: int = 16,

tests/entrypoints/llm/test_chat.py

@@ -64,7 +64,6 @@ def test_multi_chat():
 def test_chat_multi_image(image_urls: list[str]):
     llm = LLM(
         model="microsoft/Phi-3.5-vision-instruct",
-        dtype="bfloat16",
         max_model_len=4096,
         max_num_seqs=5,
         enforce_eager=True,

tests/entrypoints/openai/test_audio.py

@@ -18,8 +18,6 @@ TEST_AUDIO_URLS = [
 @pytest.fixture(scope="module")
 def server():
     args = [
-        "--dtype",
-        "bfloat16",
         "--max-model-len",
         "2048",
         "--max-num-seqs",

tests/entrypoints/openai/test_video.py

@@ -24,8 +24,6 @@ def server():
     args = [
         "--task",
         "generate",
-        "--dtype",
-        "bfloat16",
         "--max-model-len",
         "32768",
         "--max-num-seqs",

tests/entrypoints/openai/test_vision.py

@@ -25,8 +25,6 @@ def server():
     args = [
         "--task",
         "generate",
-        "--dtype",
-        "bfloat16",
         "--max-model-len",
         "2048",
         "--max-num-seqs",

tests/entrypoints/openai/test_vision_embedding.py

@@ -28,8 +28,6 @@ def server():
     args = [
         "--task",
         "embed",
-        "--dtype",
-        "bfloat16",
         "--max-model-len",
         "2048",
         "--max-num-seqs",

tests/entrypoints/test_chat_utils.py

@@ -34,7 +34,7 @@ def phi3v_model_config():
                        tokenizer=PHI3V_MODEL_ID,
                        tokenizer_mode="auto",
                        trust_remote_code=True,
-                       dtype="bfloat16",
+                       dtype="auto",
                        seed=0,
                        limit_mm_per_prompt={
                            "image": 2,
@@ -58,7 +58,7 @@ def mllama_model_config():
                        tokenizer=MLLAMA_MODEL_ID,
                        tokenizer_mode="auto",
                        trust_remote_code=True,
-                       dtype="bfloat16",
+                       dtype="auto",
                        seed=0,
                        limit_mm_per_prompt={
                            "image": 2,
@@ -669,7 +669,7 @@ def test_multimodal_image_parsing_matches_hf(model, image_url):
                                tokenizer=MLLAMA_MODEL_ID,
                                tokenizer_mode="auto",
                                trust_remote_code=True,
-                               dtype="bfloat16",
+                               dtype="auto",
                                seed=0,
                                limit_mm_per_prompt={
                                    "image": 2,

tests/models/decoder_only/audio_language/test_ultravox.py

@@ -5,11 +5,10 @@ from typing import Optional
 import numpy as np
 import pytest
 import pytest_asyncio
-from transformers import AutoModel, AutoTokenizer, BatchEncoding
+from transformers import AutoModel, AutoTokenizer
 
 from vllm.multimodal.audio import resample_audio
 from vllm.sequence import SampleLogprobs
-from vllm.utils import STR_DTYPE_TO_TORCH_DTYPE
 
 from ....conftest import HfRunner, VllmRunner
 from ....utils import RemoteOpenAIServer
@@ -107,8 +106,6 @@ def run_test(
     **kwargs,
 ):
     """Inference result should be the same between hf and vllm."""
-    torch_dtype = STR_DTYPE_TO_TORCH_DTYPE[dtype]
-
     # NOTE: take care of the order. run vLLM first, and then run HF.
     # vLLM needs a fresh new process without cuda initialization.
     # if we run HF first, the cuda initialization will be done and it
@@ -124,15 +121,7 @@ def run_test(
             for vllm_prompt, _, audio in prompts_and_audios
         ]
 
-    def process(hf_inputs: BatchEncoding, **kwargs):
+    with hf_runner(model, dtype=dtype, auto_cls=AutoModel) as hf_model:
-        hf_inputs["audio_values"] = hf_inputs["audio_values"] \
-            .to(torch_dtype)  # type: ignore
-        return hf_inputs
-
-    with hf_runner(model,
-                   dtype=dtype,
-                   postprocess_inputs=process,
-                   auto_cls=AutoModel) as hf_model:
         hf_outputs_per_audio = [
             hf_model.generate_greedy_logprobs_limit(
                 [hf_prompt],

tests/models/decoder_only/vision_language/test_models.py

@@ -9,7 +9,7 @@ from pathlib import PosixPath
 
 import pytest
 from packaging.version import Version
-from transformers import AutoModelForPreTraining, AutoModelForVision2Seq
+from transformers import AutoModelForImageTextToText, AutoModelForVision2Seq
 from transformers import __version__ as TRANSFORMERS_VERSION
 
 from vllm.platforms import current_platform
@@ -101,7 +101,7 @@ VLM_TEST_SETTINGS = {
         prompt_formatter=lambda img_prompt: f"USER: {img_prompt}\nASSISTANT:",
         convert_assets_to_embeddings=model_utils.get_llava_embeddings,
         max_model_len=4096,
-        auto_cls=AutoModelForVision2Seq,
+        auto_cls=AutoModelForImageTextToText,
         vllm_output_post_proc=model_utils.llava_image_vllm_to_hf_output,
         custom_test_opts=[CustomTestOptions(
             inputs=custom_inputs.multi_image_multi_aspect_ratio_inputs(
@@ -121,10 +121,7 @@ VLM_TEST_SETTINGS = {
             "stop_sign": "caption es",
             "cherry_blossom": "What is in the picture?",
         }),
-        auto_cls=AutoModelForVision2Seq,
+        auto_cls=AutoModelForImageTextToText,
-        postprocess_inputs=model_utils.cast_dtype_post_processor(
-            "pixel_values"
-        ),
         vllm_output_post_proc=model_utils.paligemma_vllm_to_hf_output,
         dtype="bfloat16",
         marks=[pytest.mark.skip(reason="vLLM does not support PrefixLM attention mask")],  # noqa: E501
@@ -179,7 +176,6 @@ VLM_TEST_SETTINGS = {
     #         "cherry_blossom": "<vlm_image>Please infer the season with reason.",  # noqa: E501
     #     }),
     #     multi_image_prompt="<vlm_image><vlm_image>Describe the two images shortly.",    # noqa: E501
-    #     postprocess_inputs=model_utils.cast_dtype_post_processor("pixel_values"), # noqa: E501
     #     stop_str=["<|im_end|>"],
     #     image_size_factors=[(0.10, 0.15)],
     #     max_tokens=64,
@@ -190,7 +186,7 @@ VLM_TEST_SETTINGS = {
         test_type=VLMTestType.IMAGE,
         prompt_formatter=lambda img_prompt: f"Question: {img_prompt} Answer:",
         img_idx_to_prompt=lambda idx: "",
-        auto_cls=AutoModelForVision2Seq,
+        auto_cls=AutoModelForImageTextToText,
         vllm_output_post_proc=model_utils.blip2_vllm_to_hf_output,
     ),
     "chameleon": VLMTestInfo(
@@ -199,10 +195,7 @@ VLM_TEST_SETTINGS = {
         prompt_formatter=lambda img_prompt: f"USER: {img_prompt}\nASSISTANT:",
         max_model_len=4096,
         max_num_seqs=2,
-        auto_cls=AutoModelForVision2Seq,
+        auto_cls=AutoModelForImageTextToText,
-        postprocess_inputs=model_utils.cast_dtype_post_processor(
-            "pixel_values"
-        ),
         # For chameleon, we only compare the sequences
         vllm_output_post_proc = lambda vllm_output, model: vllm_output[:2],
         hf_output_post_proc = lambda hf_output, model: hf_output[:2],
@@ -222,7 +215,6 @@ VLM_TEST_SETTINGS = {
         }),
         multi_image_prompt="image_1:<image>\nimage_2:<image>\nWhich image can we see the car and the tower?",    # noqa: E501
         patch_hf_runner=model_utils.deepseekvl2_patch_hf_runner,
-        postprocess_inputs=model_utils.cast_dtype_post_processor("images"),
         hf_output_post_proc=model_utils.deepseekvl2_trunc_hf_output,
         stop_str=["<｜end▁of▁sentence｜>", "<｜begin▁of▁sentence｜>"],  # noqa: E501
         image_size_factors=[(), (1.0, ), (1.0, 1.0, 1.0), (0.1, 0.5, 1.0)],
@@ -240,6 +232,7 @@ VLM_TEST_SETTINGS = {
         img_idx_to_prompt=lambda idx: "",
         max_model_len=2048,
         max_num_seqs=2,
+        auto_cls=AutoModelForImageTextToText,
         use_tokenizer_eos=True,
         vllm_output_post_proc=model_utils.fuyu_vllm_to_hf_output,
         num_logprobs=10,
@@ -256,9 +249,7 @@ VLM_TEST_SETTINGS = {
         multi_image_prompt="<start_of_image><start_of_image>Describe the two images in detail.",  # noqa: E501
         max_model_len=4096,
         max_num_seqs=2,
-        # TODO: Use AutoModelForVision2Seq once transformers supports this
+        auto_cls=AutoModelForImageTextToText,
-        auto_cls=AutoModelForPreTraining,
-        dtype="bfloat16",
         vllm_runner_kwargs={"mm_processor_kwargs": {"do_pan_and_scan": True}},
         patch_hf_runner=model_utils.gemma3_patch_hf_runner,
     ),
@@ -272,7 +263,6 @@ VLM_TEST_SETTINGS = {
         }),
         max_model_len=2048,
         max_num_seqs=2,
-        dtype="bfloat16",
         get_stop_token_ids=lambda tok: [151329, 151336, 151338],
         patch_hf_runner=model_utils.glm4v_patch_hf_runner,
         # The image embeddings match with HF but the outputs of the language
@@ -295,7 +285,6 @@ VLM_TEST_SETTINGS = {
         }),
         multi_image_prompt="Image-1: <image>\nImage-2: <image>\nDescribe the two images in short.",  # noqa: E501
         max_model_len=8192,
-        dtype="bfloat16",
         use_tokenizer_eos=True,
         num_logprobs=10,
         patch_hf_runner=model_utils.h2ovl_patch_hf_runner,
@@ -307,7 +296,7 @@ VLM_TEST_SETTINGS = {
         img_idx_to_prompt=lambda idx: "<image>",
         max_model_len=8192,
         max_num_seqs=2,
-        auto_cls=AutoModelForVision2Seq,
+        auto_cls=AutoModelForImageTextToText,
         hf_output_post_proc=model_utils.idefics3_trunc_hf_output,
     ),
     "intern_vl": VLMTestInfo(
@@ -324,10 +313,6 @@ VLM_TEST_SETTINGS = {
         }),
         multi_image_prompt="Image-1: <image>\nImage-2: <image>\nDescribe the two images in short.",  # noqa: E501
         max_model_len=4096,
-        # NOTE: Mono-InternVL-2B doesn't work with fp16,
-        # it will result NaN during inference.
-        # See: https://huggingface.co/OpenGVLab/Mono-InternVL-2B/discussions/9
-        dtype="bfloat16",
         use_tokenizer_eos=True,
         patch_hf_runner=model_utils.internvl_patch_hf_runner,
     ),
@@ -336,7 +321,7 @@ VLM_TEST_SETTINGS = {
         test_type=(VLMTestType.IMAGE, VLMTestType.CUSTOM_INPUTS),
         prompt_formatter=lambda img_prompt: f"[INST] {img_prompt} [/INST]",
         max_model_len=10240,
-        auto_cls=AutoModelForVision2Seq,
+        auto_cls=AutoModelForImageTextToText,
         vllm_output_post_proc=model_utils.llava_image_vllm_to_hf_output,
         custom_test_opts=[CustomTestOptions(
             inputs=custom_inputs.multi_image_multi_aspect_ratio_inputs(
@@ -351,9 +336,6 @@ VLM_TEST_SETTINGS = {
         prompt_formatter=lambda vid_prompt: f"<|im_start|>user\n{vid_prompt}<|im_end|>\n<|im_start|>assistant\n",   # noqa: E501
         num_video_frames=16,
         max_model_len=16384,
-        postprocess_inputs=model_utils.cast_dtype_post_processor(
-            "pixel_values_videos"
-        ),
         auto_cls=AutoModelForVision2Seq,
         vllm_output_post_proc=model_utils.llava_onevision_vllm_to_hf_output,
         custom_test_opts=[CustomTestOptions(
@@ -378,11 +360,8 @@ VLM_TEST_SETTINGS = {
         test_type=(VLMTestType.IMAGE, VLMTestType.MULTI_IMAGE),
         prompt_formatter=lambda img_prompt: f"<|start_header_id|>user<|end_header_id|>\n\n{img_prompt}<|eot_id|><|start_header_id|>assistant<|end_header_id|>\n\n",  # noqa: E501
         max_model_len=4096,
-        postprocess_inputs=model_utils.cast_dtype_post_processor(
-            "pixel_values"
-        ),
         get_stop_token_ids=lambda tok: [128009],
-        auto_cls=AutoModelForVision2Seq,
+        auto_cls=AutoModelForImageTextToText,
         vllm_output_post_proc=model_utils.mantis_vllm_to_hf_output,
         patch_hf_runner=model_utils.mantis_patch_hf_runner,
         marks=[
@@ -400,8 +379,8 @@ VLM_TEST_SETTINGS = {
         max_model_len=4096,
         max_num_seqs=2,
         get_stop_token_ids=lambda tok: [tok.eos_id, tok.eot_id],
-        postprocess_inputs=model_utils.wrap_inputs_post_processor,
         hf_output_post_proc=model_utils.minicpmv_trunc_hf_output,
+        patch_hf_runner=model_utils.minicpmv_25_patch_hf_runner,
     ),
     "minicpmo_26": VLMTestInfo(
         models=["openbmb/MiniCPM-o-2_6"],
@@ -411,11 +390,8 @@ VLM_TEST_SETTINGS = {
         max_model_len=4096,
         max_num_seqs=2,
         get_stop_token_ids=lambda tok: tok.convert_tokens_to_ids(['<|im_end|>', '<|endoftext|>']),  # noqa: E501
-        postprocess_inputs=model_utils.ignore_inputs_post_processor(
-            "image_sizes"
-        ),
         hf_output_post_proc=model_utils.minicpmv_trunc_hf_output,
-        patch_hf_runner=model_utils.minicpmo_patch_hf_runner
+        patch_hf_runner=model_utils.minicpmo_26_patch_hf_runner,
     ),
     "minicpmv_26": VLMTestInfo(
         models=["openbmb/MiniCPM-V-2_6"],
@@ -425,10 +401,8 @@ VLM_TEST_SETTINGS = {
         max_model_len=4096,
         max_num_seqs=2,
         get_stop_token_ids=lambda tok: tok.convert_tokens_to_ids(['<|im_end|>', '<|endoftext|>']),  # noqa: E501
-        postprocess_inputs=model_utils.ignore_inputs_post_processor(
-            "image_sizes"
-        ),
         hf_output_post_proc=model_utils.minicpmv_trunc_hf_output,
+        patch_hf_runner=model_utils.minicpmv_26_patch_hf_runner,
     ),
     "molmo": VLMTestInfo(
         models=["allenai/Molmo-7B-D-0924"],
@@ -437,7 +411,6 @@ VLM_TEST_SETTINGS = {
         max_model_len=4096,
         max_num_seqs=2,
         patch_hf_runner=model_utils.molmo_patch_hf_runner,
-        postprocess_inputs=model_utils.molmo_post_processor,
     ),
     # Tests for phi3v currently live in another file because of a bug in
     # transformers. Once this issue is fixed, we can enable them here instead.
@@ -463,7 +436,7 @@ VLM_TEST_SETTINGS = {
         img_idx_to_prompt=lambda idx: "[IMG]",
         max_model_len=8192,
         max_num_seqs=2,
-        auto_cls=AutoModelForVision2Seq,
+        auto_cls=AutoModelForImageTextToText,
         marks=[large_gpu_mark(min_gb=48)],
     ),
     "qwen_vl": VLMTestInfo(
@@ -481,10 +454,7 @@ VLM_TEST_SETTINGS = {
         models=["facebook/chameleon-7b"],
         prompt_formatter=lambda img_prompt: f"USER: {img_prompt}\nASSISTANT:",
         max_model_len=4096,
-        auto_cls=AutoModelForVision2Seq,
+        auto_cls=AutoModelForImageTextToText,
-        postprocess_inputs=model_utils.cast_dtype_post_processor(
-            "pixel_values"
-        ),
         vllm_output_post_proc = lambda vllm_output, model: vllm_output[:2],
         hf_output_post_proc = lambda hf_output, model: hf_output[:2],
         comparator=check_outputs_equal,
@@ -495,7 +465,7 @@ VLM_TEST_SETTINGS = {
         models=["llava-hf/llava-1.5-7b-hf"],
         prompt_formatter=lambda img_prompt: f"USER: {img_prompt}\nASSISTANT:",
         max_model_len=4096,
-        auto_cls=AutoModelForVision2Seq,
+        auto_cls=AutoModelForImageTextToText,
         vllm_output_post_proc=model_utils.llava_image_vllm_to_hf_output,
         marks=multi_gpu_marks(num_gpus=2),
         **COMMON_BROADCAST_SETTINGS # type: ignore
@@ -504,7 +474,7 @@ VLM_TEST_SETTINGS = {
         models=["llava-hf/llava-v1.6-mistral-7b-hf"],
         prompt_formatter=lambda img_prompt: f"[INST] {img_prompt} [/INST]",
         max_model_len=10240,
-        auto_cls=AutoModelForVision2Seq,
+        auto_cls=AutoModelForImageTextToText,
         vllm_output_post_proc=model_utils.llava_image_vllm_to_hf_output,
         marks=multi_gpu_marks(num_gpus=2),
         **COMMON_BROADCAST_SETTINGS # type: ignore
@@ -529,9 +499,6 @@ VLM_TEST_SETTINGS = {
         test_type=VLMTestType.CUSTOM_INPUTS,
         max_model_len=16384,
         max_num_seqs=2,
-        postprocess_inputs=model_utils.cast_dtype_post_processor(
-            "pixel_values"
-        ),
         auto_cls=AutoModelForVision2Seq,
         vllm_output_post_proc=model_utils.llava_onevision_vllm_to_hf_output,
         custom_test_opts=[CustomTestOptions(

tests/models/decoder_only/vision_language/test_pixtral.py

@@ -4,7 +4,6 @@
 Run `pytest tests/models/test_mistral.py`.
 """
 import json
-import uuid
 from dataclasses import asdict
 from typing import TYPE_CHECKING, Any, Optional
 
@@ -16,8 +15,7 @@ from mistral_common.tokens.tokenizers.mistral import MistralTokenizer
 from mistral_common.tokens.tokenizers.multimodal import image_from_chunk
 from transformers import AutoProcessor
 
-from vllm import (EngineArgs, LLMEngine, RequestOutput, SamplingParams,
+from vllm import RequestOutput, SamplingParams, TextPrompt, TokensPrompt
-                  TextPrompt, TokensPrompt)
 from vllm.multimodal import MultiModalDataBuiltins
 from vllm.multimodal.inputs import PlaceholderRange
 from vllm.sequence import Logprob, SampleLogprobs
@@ -28,7 +26,11 @@ from ...utils import check_logprobs_close
 if TYPE_CHECKING:
     from _typeshed import StrPath
 
-MODELS = ["mistralai/Pixtral-12B-2409"]
+PIXTRAL_ID = "mistralai/Pixtral-12B-2409"
+MISTRAL_SMALL_3_1_ID = "mistralai/Mistral-Small-3.1-24B-Instruct-2503"
+
+MODELS = [PIXTRAL_ID, MISTRAL_SMALL_3_1_ID]
+
 IMG_URLS = [
     "https://picsum.photos/id/237/400/300",
     "https://picsum.photos/id/231/200/300",
@@ -125,8 +127,10 @@ MAX_MODEL_LEN = [8192, 65536]
 FIXTURES_PATH = VLLM_PATH / "tests/models/fixtures"
 assert FIXTURES_PATH.exists()
 
-FIXTURE_LOGPROBS_CHAT = FIXTURES_PATH / "pixtral_chat.json"
+FIXTURE_LOGPROBS_CHAT = {
-FIXTURE_LOGPROBS_ENGINE = FIXTURES_PATH / "pixtral_chat_engine.json"
+    PIXTRAL_ID: FIXTURES_PATH / "pixtral_chat.json",
+    MISTRAL_SMALL_3_1_ID: FIXTURES_PATH / "mistral_small_3_chat.json",
+}
 
 OutputsLogprobs = list[tuple[list[int], str, Optional[SampleLogprobs]]]
 
@@ -166,12 +170,12 @@ def test_chat(
     model: str,
     dtype: str,
 ) -> None:
-    EXPECTED_CHAT_LOGPROBS = load_outputs_w_logprobs(FIXTURE_LOGPROBS_CHAT)
+    EXPECTED_CHAT_LOGPROBS = load_outputs_w_logprobs(
+        FIXTURE_LOGPROBS_CHAT[model])
     with vllm_runner(
             model,
             dtype=dtype,
             tokenizer_mode="mistral",
-            enable_chunked_prefill=False,
             max_model_len=max_model_len,
             limit_mm_per_prompt=LIMIT_MM_PER_PROMPT,
     ) as vllm_model:
@@ -183,70 +187,40 @@ def test_chat(
             outputs.extend(output)
 
     logprobs = vllm_runner._final_steps_generate_w_logprobs(outputs)
+    # Remove last `None` prompt_logprobs to compare with fixture
+    for i in range(len(logprobs)):
+        assert logprobs[i][-1] is None
+        logprobs[i] = logprobs[i][:-1]
     check_logprobs_close(outputs_0_lst=EXPECTED_CHAT_LOGPROBS,
                          outputs_1_lst=logprobs,
                          name_0="h100_ref",
                          name_1="output")
 
 
-@large_gpu_test(min_gb=80)
-@pytest.mark.parametrize("model", MODELS)
-@pytest.mark.parametrize("dtype", ["bfloat16"])
-def test_model_engine(vllm_runner, model: str, dtype: str) -> None:
-    EXPECTED_ENGINE_LOGPROBS = load_outputs_w_logprobs(FIXTURE_LOGPROBS_ENGINE)
-    args = EngineArgs(
-        model=model,
-        tokenizer_mode="mistral",
-        enable_chunked_prefill=False,
-        limit_mm_per_prompt=LIMIT_MM_PER_PROMPT,
-        dtype=dtype,
-    )
-    engine = LLMEngine.from_engine_args(args)
-
-    engine.add_request(uuid.uuid4().hex, ENGINE_INPUTS[0], SAMPLING_PARAMS)
-    engine.add_request(uuid.uuid4().hex, ENGINE_INPUTS[1], SAMPLING_PARAMS)
-
-    outputs = []
-    count = 0
-    while True:
-        out = engine.step()
-        count += 1
-        for request_output in out:
-            if request_output.finished:
-                outputs.append(request_output)
-
-        if count == 2:
-            engine.add_request(uuid.uuid4().hex, ENGINE_INPUTS[2],
-                               SAMPLING_PARAMS)
-        if not engine.has_unfinished_requests():
-            break
-
-    logprobs = vllm_runner._final_steps_generate_w_logprobs(outputs)
-    check_logprobs_close(outputs_0_lst=EXPECTED_ENGINE_LOGPROBS,
-                         outputs_1_lst=logprobs,
-                         name_0="h100_ref",
-                         name_1="output")
-
-
 @large_gpu_test(min_gb=48)
 @pytest.mark.parametrize(
     "prompt,expected_ranges",
     [(_create_engine_inputs_hf(IMG_URLS[:1]), [{
-        "offset": 10,
+        "offset": 11,
         "length": 494
     }]),
      (_create_engine_inputs_hf(IMG_URLS[1:4]), [{
-         "offset": 10,
+         "offset": 11,
          "length": 266
      }, {
-         "offset": 276,
+         "offset": 277,
          "length": 1056
      }, {
-         "offset": 1332,
+         "offset": 1333,
          "length": 418
      }])])
-def test_multi_modal_placeholders(
+def test_multi_modal_placeholders(vllm_runner, prompt,
-        vllm_runner, prompt, expected_ranges: list[PlaceholderRange]) -> None:
+                                  expected_ranges: list[PlaceholderRange],
+                                  monkeypatch) -> None:
+
+    # This placeholder checking test only works with V0 engine
+    # where `multi_modal_placeholders` is returned with `RequestOutput`
+    monkeypatch.setenv("VLLM_USE_V1", "0")
     with vllm_runner(
             "mistral-community/pixtral-12b",
             max_model_len=8192,

tests/models/decoder_only/vision_language/vlm_utils/core.py

@@ -4,7 +4,6 @@ from typing import Any, Callable, Optional, Union
 
 import torch
 from PIL.Image import Image
-from transformers import BatchEncoding
 from transformers.models.auto.auto_factory import _BaseAutoModelClass
 
 from vllm.config import TaskOption
@@ -31,7 +30,6 @@ def run_test(
     vllm_output_post_proc: Optional[Callable[[RunnerOutput, str], Any]],
     auto_cls: type[_BaseAutoModelClass],
     use_tokenizer_eos: bool,
-    postprocess_inputs: Callable[[BatchEncoding], BatchEncoding],
     comparator: Callable[..., None],
     get_stop_token_ids: Optional[Callable[[AnyTokenizer], list[int]]],
     stop_str: Optional[list[str]],
@@ -101,7 +99,6 @@ def run_test(
     hf_model = hf_runner(model,
                          dtype=dtype,
                          auto_cls=auto_cls,
-                         postprocess_inputs=postprocess_inputs,
                          model_kwargs=hf_model_kwargs)
 
     # Some models need to patch things like the model processor, e.g., internvl

tests/models/decoder_only/vision_language/vlm_utils/model_utils.py

@@ -6,16 +6,15 @@ typically specific to a small subset of models.
 import re
 import types
 from pathlib import PosixPath
-from typing import Callable, Optional, Union
+from typing import Optional, Union
 
 import torch
 from PIL.Image import Image
-from transformers import (AutoConfig, AutoTokenizer, BatchEncoding,
+from transformers import (AutoConfig, AutoTokenizer, BatchFeature,
                           GenerationConfig)
 
 from vllm.sequence import SampleLogprobs
 from vllm.transformers_utils.tokenizer import patch_padding_side
-from vllm.utils import STR_DTYPE_TO_TORCH_DTYPE
 
 from .....conftest import HfRunner, ImageAsset, _ImageAssets
 from .types import RunnerOutput
@@ -211,40 +210,6 @@ def get_llava_embeddings(image_assets: _ImageAssets):
     return [asset.image_embeds for asset in image_assets]
 
 
-####### postprocessors to run on HF BatchEncoding
-def cast_dtype_post_processor(
-        hf_inp_key: str) -> Callable[[BatchEncoding, str], BatchEncoding]:
-    """Gets a handle to a post processor which converts a given key into a
-    target data type."""
-
-    def process(hf_inputs: BatchEncoding, dtype: str):
-        torch_dtype = STR_DTYPE_TO_TORCH_DTYPE[dtype]
-        hf_inputs[hf_inp_key] = hf_inputs[hf_inp_key].to(torch_dtype)
-        return hf_inputs
-
-    return process
-
-
-def ignore_inputs_post_processor(
-        hf_inp_key: str) -> Callable[[BatchEncoding, str], BatchEncoding]:
-    """Gets a handle to a post processor which ignores a given key."""
-
-    def process(hf_inputs: BatchEncoding, dtype: str):
-        del hf_inputs[hf_inp_key]
-        return hf_inputs
-
-    return process
-
-
-def wrap_inputs_post_processor(hf_inputs: BatchEncoding, dtype: str):
-    return {"model_inputs": hf_inputs}
-
-
-def molmo_post_processor(hf_inputs: BatchEncoding, dtype: str):
-    hf_inputs = cast_dtype_post_processor("images")(hf_inputs, dtype)
-    return {k: v.unsqueeze(0) for k, v in hf_inputs.items()}
-
-
 ####### Prompt path encoders for models that need models on disk
 def qwen_prompt_path_encoder(
         tmp_path: PosixPath, prompt: str, assets: Union[list[ImageAsset],
@@ -295,8 +260,7 @@ def deepseekvl2_patch_hf_runner(hf_model: HfRunner) -> HfRunner:
             for k in inputs.keys()  # noqa
             if k not in ("seq_lens", "sft_format")
         }
-        inputs = BatchEncoding(data=inputs, tensor_type="pt")
+        return BatchFeature(data=inputs, tensor_type="pt")
-        return inputs
 
     hf_model.processor = processor
     hf_model.model.get_output_embeddings = lambda: \
@@ -529,10 +493,52 @@ def mantis_patch_hf_runner(hf_model: HfRunner) -> HfRunner:
     return hf_model
 
 
-def minicpmo_patch_hf_runner(hf_model: HfRunner) -> HfRunner:
+def minicpmv_25_patch_hf_runner(hf_model: HfRunner) -> HfRunner:
     orig_generate = hf_model.model.generate
 
-    def _generate(self, *args, **kwargs):
+    def _generate(
+        self,
+        *args,
+        input_ids=None,
+        pixel_values=None,
+        image_sizes=None,
+        image_bound=None,
+        tgt_sizes=None,
+        **kwargs,
+    ):
+        model_inputs = {
+            "input_ids": input_ids,
+            "pixel_values": pixel_values,
+            "image_sizes": image_sizes,
+            "image_bound": image_bound,
+            "tgt_sizes": tgt_sizes,
+        }
+        for k in list(model_inputs.keys()):
+            if model_inputs[k] is None:
+                model_inputs.pop(k)
+
+        return orig_generate(model_inputs, *args, decode_text=False, **kwargs)
+
+    hf_model.model.generate = types.MethodType(_generate, hf_model.model)
+
+    return hf_model
+
+
+def minicpmo_26_patch_hf_runner(hf_model: HfRunner) -> HfRunner:
+    orig_generate = hf_model.model.generate
+
+    def _generate(self, *args, image_sizes=None, **kwargs):
+        return orig_generate(*args, decode_text=False, **kwargs)
+
+    hf_model.model.generate = types.MethodType(_generate, hf_model.model)
+
+    return hf_model
+
+
+def minicpmv_26_patch_hf_runner(hf_model: HfRunner) -> HfRunner:
+    orig_generate = hf_model.model.generate
+
+    def _generate(self, *args, image_sizes=None, **kwargs):
         return orig_generate(*args, decode_text=False, **kwargs)
 
     hf_model.model.generate = types.MethodType(_generate, hf_model.model)
@@ -551,10 +557,11 @@ def molmo_patch_hf_runner(hf_model: HfRunner) -> HfRunner:
 
     def _generate(self, max_new_tokens=None, do_sample=None, **kwargs):
         batch = {
-            k: kwargs.pop(k)
+            k: kwargs.pop(k).unsqueeze(0)
             for k in ("input_ids", "images", "image_input_idx", "image_masks")
             if k in kwargs
         }
+        batch = BatchFeature(batch).to(dtype=self.dtype)
 
         return self.generate_from_batch(
             batch,

tests/models/decoder_only/vision_language/vlm_utils/types.py

@@ -8,13 +8,12 @@ from typing import Any, Callable, NamedTuple, Optional, Union
 import torch
 from PIL.Image import Image
 from pytest import MarkDecorator
-from transformers import AutoModelForCausalLM, BatchEncoding
+from transformers import AutoModelForCausalLM
 from transformers.models.auto.auto_factory import _BaseAutoModelClass
 
 from vllm.config import TaskOption
 from vllm.sequence import SampleLogprobs
 from vllm.transformers_utils.tokenizer import AnyTokenizer
-from vllm.utils import identity
 
 from .....conftest import IMAGE_ASSETS, HfRunner, ImageAsset, _ImageAssets
 from ....utils import check_logprobs_close
@@ -110,11 +109,6 @@ class VLMTestInfo(NamedTuple):
     # Indicates we should explicitly pass the EOS from the tokenizer
     use_tokenizer_eos: bool = False
     auto_cls: type[_BaseAutoModelClass] = AutoModelForCausalLM
-    # Callable to pass to the HF runner to run on inputs; for now, we also pass
-    # the data type to input post processing, because almost all of the uses of
-    # postprocess_inputs are to fix the data types of BatchEncoding values.
-    postprocess_inputs: Callable[[BatchEncoding, str],
-                                 BatchEncoding] = identity
     patch_hf_runner: Optional[Callable[[HfRunner], HfRunner]] = None
 
     # Post processors that if defined, will run oun the outputs of the
@@ -130,7 +124,7 @@ class VLMTestInfo(NamedTuple):
     # is all combinations of .models + all fields below
     max_tokens: Union[int, tuple[int]] = 128
     num_logprobs: Union[int, tuple[int]] = 5
-    dtype: Union[str, Iterable[str]] = "half"
+    dtype: Union[str, Union[list[str], tuple[str, ...]]] = "auto"
     distributed_executor_backend: Optional[Union[str, Iterable[str]]] = None
     # Only expanded in video tests
     num_video_frames: Union[int, tuple[int]] = 16
@@ -171,7 +165,6 @@ class VLMTestInfo(NamedTuple):
             "vllm_output_post_proc": self.vllm_output_post_proc,
             "auto_cls": self.auto_cls,
             "use_tokenizer_eos": self.use_tokenizer_eos,
-            "postprocess_inputs": self.postprocess_inputs,
             "comparator": self.comparator,
             "get_stop_token_ids": self.get_stop_token_ids,
             "hf_model_kwargs": self.hf_model_kwargs,

tests/models/embedding/vision_language/test_dse_qwen2_vl.py

@@ -1,12 +1,12 @@
 # SPDX-License-Identifier: Apache-2.0
 
-from functools import partial
 from typing import Callable
 
 import pytest
 import torch
+import torch.nn.functional as F
 from PIL import Image
-from transformers import BatchEncoding, Qwen2VLForConditionalGeneration
+from transformers import Qwen2VLForConditionalGeneration
 
 from ....conftest import IMAGE_ASSETS, HfRunner, PromptImageInput, VllmRunner
 from ....utils import large_gpu_test
@@ -75,10 +75,6 @@ def apply_chat_template_and_add_eos(
     return prompt
 
 
-def postprocess_inputs(hf_model: HfRunner, inputs: BatchEncoding, **kwargs):
-    return hf_model.model.prepare_inputs_for_generation(**inputs, **kwargs)
-
-
 def _run_test(
     hf_runner: type[HfRunner],
     vllm_runner: type[VllmRunner],
@@ -118,14 +114,8 @@ def _run_test(
     with hf_runner(model,
                    dtype=dtype,
                    auto_cls=Qwen2VLForConditionalGeneration) as hf_model:
-        hf_model.postprocess_inputs = partial(
+
-            postprocess_inputs,
+        prompts = []
-            hf_model,
-            cache_position=torch.arange(
-                0,
-                1,  # 1 for batch size
-                requires_grad=False),
-            use_cache=False)
         for text, image, embed_text in zip(input_texts, input_images,
                                            embed_texts):
             # dse requires non-standard input processing
@@ -133,20 +123,34 @@ def _run_test(
             messages = get_messages(image, text, embed_text)
             prompt = apply_chat_template_and_add_eos(
                 messages, hf_model.processor.apply_chat_template)
-            inputs = hf_model.get_inputs(
+
-                prompts=[[prompt]],
+            prompts.append(prompt)
-                images=[[image]],
+
+        all_inputs = hf_model.get_inputs(
+            prompts=prompts,
+            images=input_images,
         )
+
         with torch.no_grad():
+            all_outputs = []
+            for inputs in all_inputs:
+                inputs = hf_model.model.prepare_inputs_for_generation(
+                    **inputs,
+                    cache_position=torch.arange(1),  # 1 for batch size
+                    use_cache=False,
+                )
                 outputs = hf_model.model(
-                    **hf_model.wrap_device(inputs[0],
+                    **hf_model.wrap_device(inputs),
-                                           device=hf_model.model.device.type),
                     return_dict=True,
                     output_hidden_states=True,
                 )
-                pooled_output = torch.nn.functional.normalize(
+                pooled_output = F.normalize(outputs.hidden_states[-1][0, -1],
-                    outputs.hidden_states[-1][0, -1], p=2, dim=-1)
+                                            p=2,
-            hf_outputs.append(pooled_output.tolist())
+                                            dim=-1)
+
+                all_outputs.append(pooled_output.tolist())
+
+            hf_outputs = all_outputs
 
     check_embeddings_close(
         embeddings_0_lst=hf_outputs,

tests/models/embedding/vision_language/test_llava_next.py

@@ -2,7 +2,7 @@
 
 import pytest
 import torch.nn.functional as F
-from transformers import AutoModelForVision2Seq
+from transformers import AutoModelForImageTextToText
 
 from vllm.platforms import current_platform
 
@@ -70,7 +70,7 @@ def _run_test(
         vllm_outputs = vllm_model.encode(input_texts, images=input_images)
 
     with hf_runner(model, dtype=dtype,
-                   auto_cls=AutoModelForVision2Seq) as hf_model:
+                   auto_cls=AutoModelForImageTextToText) as hf_model:
         # Patch the issue where generation_config.json is missing
         hf_model.processor.patch_size = \
             hf_model.model.config.vision_config.patch_size
@@ -86,8 +86,7 @@ def _run_test(
         for inputs in all_inputs:
             # Based on: https://huggingface.co/royokong/e5-v
             outputs = hf_model.model(
-                **hf_model.wrap_device(inputs,
+                **hf_model.wrap_device(inputs),
-                                       device=hf_model.model.device.type),
                 return_dict=True,
                 output_hidden_states=True,
             )

tests/models/embedding/vision_language/test_phi3v.py

@@ -53,8 +53,7 @@ def _run_test(
         for inputs in all_inputs:
             # Based on: https://github.com/TIGER-AI-Lab/VLM2Vec/blob/db3b951bccabba220c1f53ab46a734e50dd2fc08/src/model.py
             outputs = hf_model.model(
-                **hf_model.wrap_device(inputs,
+                **hf_model.wrap_device(inputs),
-                                       device=hf_model.model.device.type),
                 return_dict=True,
                 output_hidden_states=True,
             )

tests/models/encoder_decoder/vision_language/test_mllama.py

@@ -4,8 +4,7 @@ from typing import Optional, overload
 
 import pytest
 import torch
-from transformers import (AutoConfig, AutoModelForVision2Seq, AutoTokenizer,
+from transformers import AutoConfig, AutoModelForImageTextToText, AutoTokenizer
-                          BatchEncoding)
 
 from vllm import LLM, SamplingParams
 from vllm.attention.backends.flash_attn import FlashAttentionMetadata
@@ -227,14 +226,10 @@ def _run_test(
             for prompts, images in inputs
         ]
 
-    def process(hf_inputs: BatchEncoding, **kwargs):
-        return hf_inputs
-
     with hf_runner(model,
                    dtype=dtype,
                    model_kwargs={"device_map": "auto"},
-                   postprocess_inputs=process,
+                   auto_cls=AutoModelForImageTextToText) as hf_model:
-                   auto_cls=AutoModelForVision2Seq) as hf_model:
         hf_outputs_per_image = [
             hf_model.generate_greedy_logprobs_limit(prompts,
                                                     max_tokens,

tests/models/utils.py

@@ -2,7 +2,7 @@
 
 import warnings
 from collections.abc import Sequence
-from typing import Optional, Union
+from typing import Any, Optional, Union
 
 import torch
 
@@ -254,9 +254,9 @@ def check_logprobs_close(
 def build_model_context(
     model_id: str,
     task: TaskOption = "auto",
-    dtype: Optional[Union[str, torch.dtype]] = None,
+    dtype: Union[str, torch.dtype] = "auto",
-    mm_processor_kwargs: Optional[dict] = None,
+    mm_processor_kwargs: Optional[dict[str, Any]] = None,
-    limit_mm_per_prompt: Optional[dict] = None,
+    limit_mm_per_prompt: Optional[dict[str, int]] = None,
     disable_mm_preprocessor_cache: bool = True,
 ):
     """Creates an InputContext for a given model.
@@ -274,9 +274,6 @@ def build_model_context(
     model_info.check_available_online(on_fail="skip")
     model_info.check_transformers_version(on_fail="skip")
 
-    if dtype is None:
-        dtype = "half"
-
     model_config = ModelConfig(
         model_id,
         task=task,

tests/multimodal/test_processing.py

@@ -7,19 +7,25 @@ from unittest.mock import MagicMock
 
 import numpy as np
 import pytest
+import torch
 from transformers import ProcessorMixin
 
 from vllm.config import ModelConfig
 from vllm.multimodal import MULTIMODAL_REGISTRY
+from vllm.multimodal.inputs import (MultiModalFieldElem, MultiModalKwargs,
+                                    MultiModalKwargsItem,
+                                    MultiModalSharedField)
 # yapf conflicts with isort for this block
 # yapf: disable
 from vllm.multimodal.processing import (PlaceholderFeaturesInfo,
-                                        PromptIndexTargets, PromptInsertion,
+                                        ProcessingCache, PromptIndexTargets,
-                                        PromptReplacement, apply_text_matches,
+                                        PromptInsertion, PromptReplacement,
+                                        apply_text_matches,
                                         apply_token_matches,
                                         find_mm_placeholders,
                                         find_text_matches, find_token_matches,
-                                        iter_token_matches)
+                                        iter_token_matches,
+                                        replace_token_matches)
 # yapf: enable
 from vllm.multimodal.profiling import MultiModalProfiler
 from vllm.transformers_utils.tokenizer import (AnyTokenizer,
@@ -89,6 +95,58 @@ def test_iter_token_matches(token_ids, match_ids, expected):
     assert all(match_len == len(match_ids) for match_len in match_lens)
 
 
+# yapf: disable
+@pytest.mark.parametrize(
+    ("token_ids", "match_ids", "new_ids", "expected"),
+    [
+        ([], [], [-1], []),
+        ([], [32000], [-1], []),
+        (
+            [32000, 32000, 32000],
+            [32000],
+            [-1],
+            [-1, -1, -1],
+        ),
+        (
+            [32000, 32000, 32000],
+            [32000, 32000],
+            [-1],
+            [-1, 32000],
+        ),
+        (
+            [32000, 32000, 32000],
+            [32000, 32000, 32000],
+            [-1],
+            [-1],
+        ),
+        (
+            [9833, 28747, 32000, 32000, 32000, 9833, 28747, 32000, 32000, 918],
+            [28747, 32000],
+            [-1],
+            [9833, -1, 32000, 32000, 9833, -1, 32000, 918],
+        ),
+        (
+            [9833, 28747, 32000, 32000, 32000, 9833, 28747, 32000, 32000, 918],
+            [28747, 32000, 32000, 32000],
+            [-1],
+            [9833, -1, 9833, 28747, 32000, 32000, 918],
+        ),
+        (
+            [9833, 28747, 32000, 32000, 32000, 9833, 28747, 32000, 32000, 918],
+            [28747, 0, 32000],
+            [-1],
+            [9833, 28747, 32000, 32000, 32000, 9833, 28747, 32000, 32000, 918],
+        ),
+    ],
+)
+# yapf: enable
+def test_replace_token_matches(token_ids, match_ids, new_ids, expected):
+    result = replace_token_matches(token_ids, match_ids, new_ids)
+
+    # Manually constructed results
+    assert result == expected
+
+
 # yapf: disable
 @pytest.mark.parametrize(
     ("prompt", "target_by_key", "expected_by_key"),
@@ -837,6 +895,45 @@ def test_find_mm_placeholders(
     assert result == expected
 
 
+def _dummy_elem(modality: str, key: str, size: int):
+    return MultiModalFieldElem(
+        modality=modality,
+        key=key,
+        data=torch.empty((size, ), dtype=torch.int8),
+        field=MultiModalSharedField(1),
+    )
+
+
+def _dummy_item(modality: str, size_by_key: dict[str, int]):
+    return MultiModalKwargsItem.from_elems([
+        _dummy_elem(modality, key, size) for key, size in size_by_key.items()
+    ])
+
+
+def _dummy_kw(size_by_key_modality: dict[str, dict[str, int]]):
+    return MultiModalKwargs.from_items([
+        _dummy_item(modality, size_by_key)
+        for modality, size_by_key in size_by_key_modality.items()
+    ])
+
+
+# yapf: disable
+@pytest.mark.parametrize(
+    ("item", "expected_size"),
+    [
+        (_dummy_item("a", {"a1": 100}), 100),
+        (_dummy_item("a", {"a1": 100, "a2": 110}), 210),
+        (_dummy_kw({"a": {"a1": 100, "a2": 110}, "b": {"b1": 120, "b2": 130}}), 460),  # noqa: E501
+    ],
+)
+# yapf: enable
+def test_cache_item_size(item, expected_size):
+    cache = ProcessingCache.get_lru_cache(2048, type(item))
+    cache[""] = item
+
+    assert cache.currsize == expected_size
+
+
 @pytest.mark.parametrize("model_id", ["llava-hf/llava-v1.6-mistral-7b-hf"])
 @pytest.mark.parametrize(
     ("limit", "num_supported", "is_valid"),
@@ -853,7 +950,7 @@ def test_limit_mm_per_prompt_dummy(model_id, limit, num_supported, is_valid):
         tokenizer_mode="auto",
         trust_remote_code=False,
         seed=0,
-        dtype="half",
+        dtype="auto",
         revision=None,
         limit_mm_per_prompt=limit_mm_per_prompt,
     )
@@ -892,7 +989,7 @@ def test_limit_mm_per_prompt_apply(model_id, num_images, limit, is_valid):
         tokenizer_mode="auto",
         trust_remote_code=False,
         seed=0,
-        dtype="half",
+        dtype="auto",
         revision=None,
         limit_mm_per_prompt=limit_mm_per_prompt,
     )
@@ -965,7 +1062,7 @@ def test_hf_processor_kwargs(model_id, call_kwargs, expected_kwargs):
         tokenizer_mode="auto",
         trust_remote_code=False,
         seed=0,
-        dtype="half",
+        dtype="auto",
         revision=None,
     )
 

tests/neuron/1_core/test_prefix_prefill.py

@@ -314,7 +314,7 @@ def get_active_block_tables(block_tables, query_lens, seq_lens, block_size,
 
         # Test edge cases
         (1, 128, 16, 1024, 4, 2, 16, False),  # large decode batch
-        (16, 4, 8, 8192, 48, 1, 128, True),  # large prefill batch
+        (16, 4, 8, 1024, 4, 2, 128, True),  # large prefill batch
         (4, 12, 32, 2048, 16, 1, 32, True),  # multi-head attention (MHA)
         (4, 12, 32, 2048, 16, 16, 32, True),  # multi-query attention (MQA)
     ])

tests/quantization/test_bitsandbytes.py

@@ -15,6 +15,8 @@ from ..utils import compare_two_settings, create_new_process_for_each_test
 
 models_4bit_to_test = [
     ("facebook/opt-125m", "quantize opt model inflight"),
+    ("mistralai/Mistral-7B-Instruct-v0.3",
+     "quantize inflight model with both HF and Mistral format weights")
 ]
 
 models_pre_qaunt_4bit_to_test = [

tests/tensorizer_loader/test_tensorizer.py

@@ -166,7 +166,7 @@ def test_vllm_model_can_load_with_lora(vllm_runner, tmp_path):
     test_prompts = multilora_inference.create_test_prompts(lora_path)
 
     # Serialize model before deserializing and binding LoRA adapters
-    with vllm_runner(model_ref, ) as vllm_model:
+    with vllm_runner(model_ref) as vllm_model:
         model_path = tmp_path / (model_ref + ".tensors")
 
         vllm_model.apply_model(
@@ -208,7 +208,7 @@ def test_load_without_tensorizer_load_format(vllm_runner):
 @pytest.mark.skipif(not is_curl_installed(), reason="cURL is not installed")
 def test_openai_apiserver_with_tensorizer(vllm_runner, tmp_path):
     ## Serialize model
-    with vllm_runner(model_ref, ) as vllm_model:
+    with vllm_runner(model_ref) as vllm_model:
         model_path = tmp_path / (model_ref + ".tensors")
 
         vllm_model.apply_model(

tests/tpu/test_compilation.py

@@ -34,7 +34,9 @@ with depyf.prepare_debug(temp_dir):
 
     # disable custom dispatcher, let Dynamo takes over
     # all the control
-    llm = LLM(model="google/gemma-2b",
+    llm = LLM(model="Qwen/Qwen2.5-1.5B-Instruct",
+              max_model_len=512,
+              max_num_seqs=64,
               enforce_eager=True,
               compilation_config={"level": CompilationLevel.DYNAMO_AS_IS})
     outputs = llm.generate(prompts, sampling_params)
@@ -44,38 +46,51 @@ with depyf.prepare_debug(temp_dir):
         print(f"Prompt: {prompt!r}, Generated text: {generated_text!r}")
         assert generated_text.startswith(answer)
 
-compiled_code = sorted(
+compiled_codes = sorted(
     glob.glob(os.path.join(temp_dir, "__transformed_code*.py")))
 
-# we should only trigger Dynamo compilation three times:
+for i, compiled_code in enumerate(compiled_codes):
-# one for the profiling phase without kv cache
+    print("{} file: {}".format(i + 1, compiled_code))
-# one for the prefill phase with symbolic shapes
+
-# one for the decode phase with symbolic shapes
+# We should only trigger Dynamo compilation 4 times:
+# 1. forward pass (symbolic)
+# 2. compute_logits (symbolic)
+# 3. forward pass (shape 16)
+# 4. forward pass (shape 32)
 # and later calls should not trigger Dynamo compilation again.
-# NOTE: it might still trigger XLA compilation.
+# NOTE: It might still trigger XLA compilation.
 
-# check we have three compiled code
+# Check we have 4 compiled codes
-# this is the assumption when we use the custom dispatcher
+assert len(compiled_codes) == 4
-assert len(compiled_code) == 3
 
-# check all the compilations are as expected
+kv_cache_prefix = "kv_cache"
-compiled_fn = sorted(
+attn_prefix = "ragged_paged_attention"
+
+# Check all the compilations are as expected
+compiled_fns = sorted(
     glob.glob(os.path.join(temp_dir, "__compiled_fn*Captured*.py")))
 
-# the first compilation is the profiling phase,
+for i, compiled_fn in enumerate(compiled_fns):
-# it should not have any kv cache
+    print("{} file: {}".format(i + 1, compiled_fn))
-with open(compiled_fn[0]) as f:
-    content = f.read()
-    assert "kv_caches" not in content
 
-# the second compilation is the prefill phase,
+# The first compilation is symbolic, so it should not have any kv_caches
-# it should have kv cache and the flash_attention op
+with open(compiled_fns[0]) as f:
-with open(compiled_fn[1]) as f:
     content = f.read()
-    assert "kv_caches" in content and "torch.ops.xla.flash_attention" in content
+    assert kv_cache_prefix not in content
 
-# the third compilation is the decode phase,
+# The second compilation is symbolic, so it should not have any kv_caches
-# it should have kv cache and the paged_attention op
+with open(compiled_fns[1]) as f:
-with open(compiled_fn[2]) as f:
     content = f.read()
-    assert "kv_caches" in content and "torch.ops.xla.paged_attention" in content
+    assert kv_cache_prefix not in content
+
+# The third compilation is shape 16, so it should have kv_caches and the
+# ragged_paged_attention
+with open(compiled_fns[2]) as f:
+    content = f.read()
+    assert (kv_cache_prefix in content and attn_prefix in content)
+
+# The forth compilation is shape 32, so it should have kv_caches and the
+# ragged_paged_attention
+with open(compiled_fns[3]) as f:
+    content = f.read()
+    assert (kv_cache_prefix in content and attn_prefix in content)

tests/tpu/test_custom_dispatcher.py

@@ -14,12 +14,17 @@ from ..utils import compare_two_settings
 def test_custom_dispatcher(monkeypatch: pytest.MonkeyPatch):
     with monkeypatch.context() as m:
         m.setenv("VLLM_RPC_TIMEOUT", "30000")
-        compare_two_settings(
+        compare_two_settings("Qwen/Qwen2.5-1.5B-Instruct",
-            "google/gemma-2b",
                              arg1=[
+                                 "--max-model-len=256",
+                                 "--max-num-seqs=32",
                                  "--enforce-eager",
                                  f"-O{CompilationLevel.DYNAMO_ONCE}",
                              ],
-            arg2=["--enforce-eager", f"-O{CompilationLevel.DYNAMO_AS_IS}"],
+                             arg2=[
+                                 "--max-model-len=256", "--max-num-seqs=32",
+                                 "--enforce-eager",
+                                 f"-O{CompilationLevel.DYNAMO_AS_IS}"
+                             ],
                              env1={},
                              env2={})

tests/v1/engine/test_async_llm.py

@@ -76,21 +76,18 @@ async def generate(engine: AsyncLLM,
 
 @pytest.mark.parametrize(
     "output_kind", [RequestOutputKind.DELTA, RequestOutputKind.FINAL_ONLY])
-@pytest.mark.parametrize("engine_args_and_prompt",
+@pytest.mark.parametrize("engine_args,prompt",
                          [(TEXT_ENGINE_ARGS, TEXT_PROMPT),
                           (VISION_ENGINE_ARGS, VISION_PROMPT)])
 @pytest.mark.asyncio
-async def test_load(
+async def test_load(monkeypatch: pytest.MonkeyPatch,
-    monkeypatch: pytest.MonkeyPatch,
                     output_kind: RequestOutputKind,
-    engine_args_and_prompt: tuple[AsyncEngineArgs, PromptType],
+                    engine_args: AsyncEngineArgs, prompt: PromptType):
-):
     # TODO(rickyx): Remove monkeypatch once we have a better way to test V1
     # so that in the future when we switch, we don't have to change all the
     # tests.
     with monkeypatch.context() as m, ExitStack() as after:
         m.setenv("VLLM_USE_V1", "1")
-        engine_args, prompt = engine_args_and_prompt
 
         engine = AsyncLLM.from_engine_args(engine_args)
         after.callback(engine.shutdown)
@@ -124,18 +121,16 @@ async def test_load(
 
 @pytest.mark.parametrize(
     "output_kind", [RequestOutputKind.DELTA, RequestOutputKind.FINAL_ONLY])
-@pytest.mark.parametrize("engine_args_and_prompt",
+@pytest.mark.parametrize("engine_args,prompt",
                          [(TEXT_ENGINE_ARGS, TEXT_PROMPT),
                           (VISION_ENGINE_ARGS, VISION_PROMPT)])
 @pytest.mark.asyncio
 async def test_abort(monkeypatch: pytest.MonkeyPatch,
                      output_kind: RequestOutputKind,
-                     engine_args_and_prompt: tuple[AsyncEngineArgs,
+                     engine_args: AsyncEngineArgs, prompt: PromptType):
-                                                   PromptType]):
 
     with monkeypatch.context() as m, ExitStack() as after:
         m.setenv("VLLM_USE_V1", "1")
-        engine_args, prompt = engine_args_and_prompt
 
         engine = AsyncLLM.from_engine_args(engine_args)
         after.callback(engine.shutdown)
@@ -193,17 +188,15 @@ async def test_abort(monkeypatch: pytest.MonkeyPatch,
 
 
 @pytest.mark.parametrize("n", [1, 3])
-@pytest.mark.parametrize("engine_args_and_prompt",
+@pytest.mark.parametrize("engine_args,prompt",
                          [(TEXT_ENGINE_ARGS, TEXT_PROMPT),
                           (VISION_ENGINE_ARGS, VISION_PROMPT)])
 @pytest.mark.asyncio
-async def test_finished_flag(monkeypatch, n: int,
+async def test_finished_flag(monkeypatch: pytest.MonkeyPatch, n: int,
-                             engine_args_and_prompt: tuple[AsyncEngineArgs,
+                             engine_args: AsyncEngineArgs, prompt: PromptType):
-                                                           PromptType]):
 
     with monkeypatch.context() as m, ExitStack() as after:
         m.setenv("VLLM_USE_V1", "1")
-        engine_args, prompt = engine_args_and_prompt
 
         engine = AsyncLLM.from_engine_args(engine_args)
         after.callback(engine.shutdown)

tests/v1/engine/test_llm_engine.py

@@ -50,7 +50,7 @@ def _get_test_sampling_params(
     """Generate random sampling params for a batch."""
 
     def get_mostly_n_gt1() -> int:
-        """Mostly n \in [2,20], ~1/3 n=1"""
+        r"""Mostly n \in [2,20], ~1/3 n=1"""
         x = random.randint(0, 28)
         if x < 10:
             return 1

tests/v1/sample/test_rejection_sampler.py

@@ -6,20 +6,23 @@ import torch
 import torch.nn.functional as F
 
 from vllm.v1.sample.metadata import SamplingMetadata
-from vllm.v1.sample.rejection_sampler import INVALID_TOKEN_ID, RejectionSampler
+from vllm.v1.sample.rejection_sampler import (PLACEHOLDER_TOKEN_ID,
+                                              RejectionSampler)
+from vllm.v1.spec_decode.metadata import SpecDecodeMetadata
 
-DEVICE = "cpu"
+DEVICE = "cuda"
 
 
 @pytest.fixture
-def sampler():
+def rejection_sampler():
     return RejectionSampler()
 
 
-def create_logits_tensor(token_ids: list[list[int]],
+def create_logits_tensor(output_token_ids: list[list[int]],
                          vocab_size: int = 100) -> torch.Tensor:
     """Helper function to create logits tensor that 
        will produce desired token ids on argmax"""
+    token_ids = [tokens[:-1] for tokens in output_token_ids]
     num_total_tokens = sum(len(tokens) for tokens in token_ids)
     logits = torch.full((num_total_tokens, vocab_size), -100.0, device=DEVICE)
     start_loc = 0
@@ -32,14 +35,21 @@ def create_logits_tensor(token_ids: list[list[int]],
 
 def create_sampling_metadata(
     all_greedy: bool,
-        generators: Optional[dict[int, Any]] = None) -> SamplingMetadata:
+    temperature: Optional[torch.Tensor] = None,
+    generators: Optional[dict[int, Any]] = None,
+) -> SamplingMetadata:
     """Create a v1 sampling metadata object with all_greedy set 
         to the given value. Either all greedy or all random sampling 
         is used.
     """
     generators = generators or {}
+    if all_greedy:
+        temperature = None
+    else:
+        assert temperature is not None
+
     return SamplingMetadata(
-        temperature=torch.tensor([]),
+        temperature=temperature,
         all_greedy=all_greedy,
         all_random=not all_greedy,
         top_p=None,
@@ -61,7 +71,7 @@ def create_sampling_metadata(
 
 
 ########################### Tests for Greedy Sampling ###################
-def test_perfect_match(sampler):
+def test_perfect_match(rejection_sampler):
     """Test when output tokens perfectly match speculated tokens"""
     spec_tokens = [[1, 2, 3]]
     output_tokens = [[1, 2, 3, 4]]  # 4 is the bonus token
@@ -70,15 +80,23 @@ def test_perfect_match(sampler):
     logits = create_logits_tensor(output_tokens)
     bonus_token_tensor = torch.tensor([output_tokens[0][-1]],
                                       device=logits.device)
+    spec_decode_metadata = SpecDecodeMetadata.make_dummy(spec_tokens,
+                                                         device=logits.device)
 
-    output = sampler(spec_tokens, None, bonus_token_tensor, logits, metadata)
+    output = rejection_sampler(
+        spec_decode_metadata,
+        draft_probs=None,
+        target_logits=logits,
+        bonus_token_ids=bonus_token_tensor,
+        sampling_metadata=metadata,
+    )
     expected = torch.tensor([[1, 2, 3, 4]],
                             dtype=torch.int,
                             device=logits.device)
     assert torch.equal(output, expected)
 
 
-def test_early_mismatch(sampler):
+def test_early_mismatch(rejection_sampler):
     """Test when there's an early mismatch in tokens"""
     spec_tokens = [[1, 2, 3]]
     output_tokens = [[1, 5, 3, 4]]  # Mismatch at position 1
@@ -87,15 +105,25 @@ def test_early_mismatch(sampler):
     logits = create_logits_tensor(output_tokens)
     bonus_token_tensor = torch.tensor([output_tokens[0][-1]],
                                       device=logits.device)
-
+    spec_decode_metadata = SpecDecodeMetadata.make_dummy(spec_tokens,
-    output = sampler(spec_tokens, None, bonus_token_tensor, logits, metadata)
-    expected = torch.tensor([[1, 5, INVALID_TOKEN_ID, INVALID_TOKEN_ID]],
-                            dtype=torch.int,
                                                          device=logits.device)
+
+    output = rejection_sampler(
+        spec_decode_metadata,
+        draft_probs=None,
+        target_logits=logits,
+        bonus_token_ids=bonus_token_tensor,
+        sampling_metadata=metadata,
+    )
+    expected = torch.tensor(
+        [[1, 5, PLACEHOLDER_TOKEN_ID, PLACEHOLDER_TOKEN_ID]],
+        dtype=torch.int,
+        device=logits.device,
+    )
     assert torch.equal(output, expected)
 
 
-def test_multiple_sequences(sampler):
+def test_multiple_sequences(rejection_sampler):
     """Test handling multiple sequences of speculated tokens"""
     spec_tokens = [[1, 2], [3]]
     output_tokens = [[1, 2, 5], [3,
@@ -105,15 +133,23 @@ def test_multiple_sequences(sampler):
     logits = create_logits_tensor(output_tokens)
     bonus_token_tensor = torch.tensor(
         [output_tokens[0][-1], output_tokens[1][-1]], device=logits.device)
+    spec_decode_metadata = SpecDecodeMetadata.make_dummy(spec_tokens,
+                                                         device=logits.device)
 
-    output = sampler(spec_tokens, None, bonus_token_tensor, logits, metadata)
+    output = rejection_sampler(
-    expected = torch.tensor([[1, 2, 5], [3, 4, INVALID_TOKEN_ID]],
+        spec_decode_metadata,
+        draft_probs=None,
+        target_logits=logits,
+        bonus_token_ids=bonus_token_tensor,
+        sampling_metadata=metadata,
+    )
+    expected = torch.tensor([[1, 2, 5], [3, 4, PLACEHOLDER_TOKEN_ID]],
                             dtype=torch.int,
                             device=logits.device)
     assert torch.equal(output, expected)
 
 
-def test_single_token_sequence(sampler):
+def test_single_token_sequence(rejection_sampler):
     """Test handling sequences with single token"""
     spec_tokens = [[1]]
     output_tokens = [[1, 2]]  # Single token with bonus token 2
@@ -122,13 +158,21 @@ def test_single_token_sequence(sampler):
     logits = create_logits_tensor(output_tokens)
     bonus_token_tensor = torch.tensor([output_tokens[0][-1]],
                                       device=logits.device)
+    spec_decode_metadata = SpecDecodeMetadata.make_dummy(spec_tokens,
+                                                         device=logits.device)
 
-    output = sampler(spec_tokens, None, bonus_token_tensor, logits, metadata)
+    output = rejection_sampler(
+        spec_decode_metadata,
+        draft_probs=None,
+        target_logits=logits,
+        bonus_token_ids=bonus_token_tensor,
+        sampling_metadata=metadata,
+    )
     expected = torch.tensor([[1, 2]], dtype=torch.int, device=logits.device)
     assert torch.equal(output, expected)
 
 
-def test_empty_sequence(sampler):
+def test_empty_sequence(rejection_sampler):
     """Test handling empty sequence of speculated tokens"""
     spec_tokens: list[list[int]] = [[]]
     output_tokens = [[5]]  # Just the bonus token
@@ -137,13 +181,21 @@ def test_empty_sequence(sampler):
     logits = create_logits_tensor(output_tokens)
     bonus_token_tensor = torch.tensor([output_tokens[0][-1]],
                                       device=logits.device)
+    spec_decode_metadata = SpecDecodeMetadata.make_dummy(spec_tokens,
+                                                         device=logits.device)
 
-    output = sampler(spec_tokens, None, bonus_token_tensor, logits, metadata)
+    output = rejection_sampler(
+        spec_decode_metadata,
+        draft_probs=None,
+        target_logits=logits,
+        bonus_token_ids=bonus_token_tensor,
+        sampling_metadata=metadata,
+    )
     expected = torch.tensor([[5]], dtype=torch.int, device=logits.device)
     assert torch.equal(output, expected)
 
 
-def test_multiple_mismatches(sampler):
+def test_multiple_mismatches(rejection_sampler):
     """Test handling multiple sequences with mismatches"""
     spec_tokens = [[1, 2, 3], [4, 5, 6]]
     output_tokens = [[1, 2, 7, 6], [4, 8, 6,
@@ -153,12 +205,22 @@ def test_multiple_mismatches(sampler):
     logits = create_logits_tensor(output_tokens)
     bonus_token_tensor = torch.tensor(
         [output_tokens[0][-1], output_tokens[1][-1]], device=logits.device)
-
+    spec_decode_metadata = SpecDecodeMetadata.make_dummy(spec_tokens,
-    output = sampler(spec_tokens, None, bonus_token_tensor, logits, metadata)
-    expected = torch.tensor([[1, 2, 7, INVALID_TOKEN_ID],
-                             [4, 8, INVALID_TOKEN_ID, INVALID_TOKEN_ID]],
-                            dtype=torch.int,
                                                          device=logits.device)
+
+    output = rejection_sampler(
+        spec_decode_metadata,
+        draft_probs=None,
+        target_logits=logits,
+        bonus_token_ids=bonus_token_tensor,
+        sampling_metadata=metadata,
+    )
+    expected = torch.tensor(
+        [[1, 2, 7, PLACEHOLDER_TOKEN_ID],
+         [4, 8, PLACEHOLDER_TOKEN_ID, PLACEHOLDER_TOKEN_ID]],
+        dtype=torch.int,
+        device=logits.device,
+    )
     assert torch.equal(output, expected)
 
 
@@ -166,18 +228,27 @@ def test_multiple_mismatches(sampler):
     "spec_tokens,output_tokens,expected",
     [
         ([[1, 2]], [[1, 2, 3]], [[1, 2, 3]]),  # Perfect match with bonus
-        ([[1]], [[2, 3]], [[2, INVALID_TOKEN_ID]]),  # First mismatch
+        ([[1]], [[2, 3]], [[2, PLACEHOLDER_TOKEN_ID]]),  # First mismatch
         ([[1, 2], [3, 4]], [[1, 5, 6], [3, 4, 7]],
-         [[1, 5, INVALID_TOKEN_ID], [3, 4, 7]]),  # Mixed matches
+         [[1, 5, PLACEHOLDER_TOKEN_ID], [3, 4, 7]]),  # Mixed matches
     ])
-def test_parametrized_cases(sampler, spec_tokens, output_tokens, expected):
+def test_parametrized_cases(rejection_sampler, spec_tokens, output_tokens,
+                            expected):
     """Parametrized test for various matching scenarios"""
     metadata = create_sampling_metadata(all_greedy=True)
     logits = create_logits_tensor(output_tokens)
     bonus_token_tensor = torch.tensor([tokens[-1] for tokens in output_tokens],
                                       device=logits.device)
+    spec_decode_metadata = SpecDecodeMetadata.make_dummy(spec_tokens,
+                                                         device=logits.device)
 
-    output = sampler(spec_tokens, None, bonus_token_tensor, logits, metadata)
+    output = rejection_sampler(
+        spec_decode_metadata,
+        draft_probs=None,
+        target_logits=logits,
+        bonus_token_ids=bonus_token_tensor,
+        sampling_metadata=metadata,
+    )
     expected_tensor = torch.tensor(expected,
                                    dtype=torch.int,
                                    device=logits.device)
@@ -190,21 +261,31 @@ def test_parametrized_cases(sampler, spec_tokens, output_tokens, expected):
 @pytest.mark.parametrize("batch_size", [1, 4, 8])
 @pytest.mark.parametrize("frac_seeded", [0.0, 0.5])
 @pytest.mark.parametrize("n_rep", [20])
-def test_deterministic_when_seeded(sampler, k: int, vocab_size: int,
+def test_deterministic_when_seeded(
-                                   batch_size: int, frac_seeded: float,
+    rejection_sampler,
-                                   n_rep: int):
+    k: int,
-    draft_probs = torch.rand(batch_size, k, vocab_size, dtype=torch.float32)
+    vocab_size: int,
-    target_probs = torch.rand(batch_size * (k + 1),
+    batch_size: int,
+    frac_seeded: float,
+    n_rep: int,
+):
+    num_tokens = batch_size * k
+    draft_probs = torch.rand(num_tokens,
                              vocab_size,
-                              dtype=torch.float32)
+                             dtype=torch.float32,
+                             device=DEVICE)
+    draft_probs = F.softmax(draft_probs, dim=-1)
+    target_logits = torch.rand_like(draft_probs)
     bonus_token_ids = torch.randint(low=0,
                                     high=vocab_size,
                                     size=(batch_size, 1),
-                                    dtype=torch.int64)
+                                    dtype=torch.int64,
+                                    device=DEVICE)
     draft_token_ids = torch.randint(low=0,
                                     high=vocab_size,
                                     size=(batch_size, k),
-                                    dtype=torch.int64)
+                                    dtype=torch.int64,
+                                    device=DEVICE)
 
     seeded_mask = torch.rand(batch_size, dtype=torch.float32) <= frac_seeded
 
@@ -215,10 +296,21 @@ def test_deterministic_when_seeded(sampler, k: int, vocab_size: int,
             for i in range(batch_size) if seeded_mask[i]
         }
 
+        temperature = torch.ones(batch_size,
+                                 dtype=torch.float32,
+                                 device=DEVICE)
         sampling_metadata = create_sampling_metadata(all_greedy=False,
+                                                     temperature=temperature,
                                                      generators=seeded_seqs)
-        rep_result = sampler(draft_token_ids.tolist(), draft_probs,
+        spec_decode_metadata = SpecDecodeMetadata.make_dummy(
-                             bonus_token_ids, target_probs, sampling_metadata)
+            draft_token_ids.tolist(), device=DEVICE)
+        rep_result = rejection_sampler(
+            spec_decode_metadata,
+            draft_probs=draft_probs,
+            target_logits=target_logits,
+            bonus_token_ids=bonus_token_ids,
+            sampling_metadata=sampling_metadata,
+        )
 
         results.append(rep_result)
 
@@ -257,10 +349,10 @@ def test_rejection_sampling_approximates_target_distribution():
     num_reference_probs = 100
 
     # Prepare draft, target, and reference probability distributions
-    draft_probs, target_probs = (F.softmax(
+    draft_probs = F.softmax(torch.rand(vocab_size, dtype=torch.float32),
-        torch.rand(vocab_size, dtype=torch.float32),
+                            dim=-1)
-        dim=-1,
+    target_logits = torch.rand(vocab_size, dtype=torch.float32)
-    ) for _ in range(2))
+    target_probs = F.softmax(target_logits, dim=-1)
     reference_probs = F.softmax(
         torch.rand(num_reference_probs, vocab_size, dtype=torch.float32),
         dim=-1,
@@ -273,7 +365,7 @@ def test_rejection_sampling_approximates_target_distribution():
     for num_samples in sample_sizes:
         # Sample using rejection sampling.
         rej_sample_probs = estimate_rejection_sampling_pdf(
-            draft_probs, target_probs, k, vocab_size, num_samples)
+            draft_probs, target_logits, k, vocab_size, num_samples)
         rej_sample_probs = rej_sample_probs.to(DEVICE)
 
         # Average distance from reference probs.
@@ -313,7 +405,7 @@ def get_ratio_first_to_last(elements: list[float]) -> float:
 
 def estimate_rejection_sampling_pdf(
     draft_probs: torch.Tensor,
-    target_probs: torch.Tensor,
+    target_logits: torch.Tensor,
     k: int,
     vocab_size: int,
     num_samples: int,
@@ -323,35 +415,44 @@ def estimate_rejection_sampling_pdf(
 
     Args:
         draft_probs: Draft probability distribution.
-        target_probs: Target probability distribution.
+        target_logits: Target logits.
         num_samples: Number of samples to draw.
 
     Returns:
         Estimated probability distribution of the output tokens.
     """
-    sampler = RejectionSampler()
+    rejection_sampler = RejectionSampler()
-    # Repeat draft probs num_samples times.
+    num_tokens = num_samples * k
+    # Repeat draft probs num_samples * k times.
     draft_probs = draft_probs.reshape(1, 1,
                                       vocab_size).repeat(num_samples, k, 1)
 
-    # Repeat target probs num_samples * (k + 1) times.
+    # Repeat target probs num_tokens times.
-    target_probs = target_probs.reshape(1, 1, vocab_size).repeat(
+    target_logits = target_logits.reshape(1, vocab_size).repeat(num_tokens, 1)
-        num_samples, k + 1, 1).reshape(num_samples * (k + 1), vocab_size)
 
     # Randomly sample draft token ids from draft probs.
     draft_token_ids = torch.multinomial(draft_probs[:, 0, :],
                                         num_samples=k,
                                         replacement=True).reshape(
                                             num_samples, k)
+    draft_probs = draft_probs.view(num_tokens, vocab_size)
 
     # Bonus tokens not used but required.
     bonus_token_ids = torch.zeros((1, 1), dtype=torch.int64,
                                   device=DEVICE).repeat(num_samples, 1)
 
-    sampling_metadata = create_sampling_metadata(all_greedy=False)
+    temperature = torch.ones(num_samples, dtype=torch.float32, device=DEVICE)
-    output_token_ids = sampler(draft_token_ids.tolist(), draft_probs,
+    sampling_metadata = create_sampling_metadata(all_greedy=False,
-                               bonus_token_ids, target_probs,
+                                                 temperature=temperature)
-                               sampling_metadata)
+    spec_decode_metadata = SpecDecodeMetadata.make_dummy(
+        draft_token_ids.tolist(), device=bonus_token_ids.device)
+    output_token_ids = rejection_sampler(
+        spec_decode_metadata,
+        draft_probs=draft_probs,
+        target_logits=target_logits,
+        bonus_token_ids=bonus_token_ids,
+        sampling_metadata=sampling_metadata,
+    )
     output_token_ids = output_token_ids[:, :-1].flatten()
 
     hist = torch.histogram(output_token_ids.to(dtype=torch.float,

tests/v1/tpu/test_basic.py

@@ -15,9 +15,10 @@ if TYPE_CHECKING:
     from tests.conftest import VllmRunner
 
 MODELS = [
+    "Qwen/Qwen2.5-1.5B-Instruct",
+    # TODO: Enable this models with v6e
     # "Qwen/Qwen2-7B-Instruct",
-    "meta-llama/Llama-3.1-8B",
+    # "meta-llama/Llama-3.1-8B",
-    # TODO: Add models here as necessary
 ]
 
 TENSOR_PARALLEL_SIZES = [1]

vllm/config.py

@@ -347,7 +347,7 @@ class ModelConfig:
         self.encoder_config = self._get_encoder_config()
         self.hf_image_processor_config = get_hf_image_processor_config(
             self.model, revision)
-        self.dtype = _get_and_verify_dtype(self.hf_text_config, dtype)
+        self.dtype = _get_and_verify_dtype(self.hf_config, dtype)
         self.use_async_output_proc = use_async_output_proc
         self.mm_processor_kwargs = mm_processor_kwargs
         self.disable_mm_preprocessor_cache = disable_mm_preprocessor_cache
@@ -2526,6 +2526,14 @@ def _get_and_verify_dtype(
     # NOTE: getattr(config, "torch_dtype", torch.float32) is not correct
     # because config.torch_dtype can be None.
     config_dtype = getattr(config, "torch_dtype", None)
+
+    # Fallbacks for multi-modal models if the root config
+    # does not define torch_dtype
+    if config_dtype is None and hasattr(config, "text_config"):
+        config_dtype = getattr(config.text_config, "torch_dtype", None)
+    if config_dtype is None and hasattr(config, "vision_config"):
+        config_dtype = getattr(config.vision_config, "torch_dtype", None)
+
     if config_dtype is None:
         config_dtype = torch.float32
 
@@ -2533,15 +2541,7 @@ def _get_and_verify_dtype(
         dtype = dtype.lower()
         if dtype == "auto":
             if config_dtype == torch.float32:
-                if config.model_type in ("gemma2", "gemma3", "gemma3_text"):
+                # Following common practice, we use float16 for float32 models
-                    logger.info(
-                        "For Gemma 2 and 3, we downcast float32 to bfloat16 "
-                        "instead of float16 by default. Please specify `dtype` "
-                        "if you want to use float16.")
-                    torch_dtype = torch.bfloat16
-                else:
-                    # Following the common practice, we use float16 for float32
-                    # models.
                 torch_dtype = torch.float16
             else:
                 torch_dtype = config_dtype

vllm/engine/arg_utils.py

@@ -1469,8 +1469,12 @@ class EngineArgs:
             return False
 
         # Need at least Ampere for now (FA support required).
+        # Skip this check if we are running on a non-GPU platform,
+        # or if the device capability is not available
+        # (e.g. in a Ray actor without GPUs).
         from vllm.platforms import current_platform
         if (current_platform.is_cuda()
+                and current_platform.get_device_capability()
                 and current_platform.get_device_capability().major < 8):
             _raise_or_fallback(feature_name="Compute Capability < 8.0",
                                recommend_to_remove=False)
@@ -1574,6 +1578,13 @@ class EngineArgs:
             _raise_or_fallback(feature_name=name, recommend_to_remove=True)
             return False
 
+        # No support for device type other than CUDA, AMD (experiemntal) or
+        # TPU (experimental) so far.
+        if not (current_platform.is_cuda_alike() or current_platform.is_tpu()):
+            _raise_or_fallback(
+                feature_name=f"device type={current_platform.device_type}",
+                recommend_to_remove=False)
+            return False
         #############################################################
         # Experimental Features - allow users to opt in.
 

vllm/entrypoints/openai/protocol.py

@@ -548,7 +548,7 @@ class ChatCompletionRequest(OpenAIBaseModel):
             if top_logprobs < 0:
                 raise ValueError("`top_logprobs` must be a positive value.")
 
-            if not data.get("logprobs"):
+            if top_logprobs > 0 and not data.get("logprobs"):
                 raise ValueError(
                     "when using `top_logprobs`, `logprobs` must be set to true."
                 )

vllm/envs.py

@@ -35,7 +35,6 @@ if TYPE_CHECKING:
     VLLM_TRACE_FUNCTION: int = 0
     VLLM_ATTENTION_BACKEND: Optional[str] = None
     VLLM_USE_FLASHINFER_SAMPLER: Optional[bool] = None
-    VLLM_USE_FLASHINFER_REJECTION_SAMPLER: bool = False
     VLLM_FLASHINFER_FORCE_TENSOR_CORES: bool = False
     VLLM_PP_LAYER_PARTITION: Optional[str] = None
     VLLM_CPU_KVCACHE_SPACE: int = 0

vllm/executor/multiproc_worker_utils.py

@@ -16,12 +16,8 @@ import torch
 
 from vllm.config import VllmConfig
 from vllm.logger import init_logger
-from vllm.triton_utils.importing import HAS_TRITON
 from vllm.utils import _check_multiproc_method, get_mp_context, run_method
 
-if HAS_TRITON:
-    from vllm.triton_utils import maybe_set_triton_cache_manager
-
 logger = init_logger(__name__)
 
 T = TypeVar('T')
@@ -314,7 +310,3 @@ def set_multiprocessing_worker_envs(parallel_config):
             current_parallelism, default_omp_num_threads)
         os.environ["OMP_NUM_THREADS"] = str(default_omp_num_threads)
         torch.set_num_threads(default_omp_num_threads)
-
-    # workaround for https://github.com/vllm-project/vllm/issues/6103
-    if HAS_TRITON and parallel_config.world_size > 1:
-        maybe_set_triton_cache_manager()

vllm/lora/models.py

@@ -30,6 +30,7 @@ from vllm.lora.utils import (from_layer, from_layer_logits_processor,
                              is_regex_target_modules,
                              parse_fine_tuned_lora_name, replace_submodule)
 from vllm.model_executor.models import SupportsLoRA, supports_multimodal
+from vllm.model_executor.models.interfaces import is_pooling_model
 from vllm.model_executor.models.module_mapping import MultiModelKeys
 from vllm.model_executor.models.utils import PPMissingLayer, WeightsMapper
 from vllm.utils import is_pin_memory_available
@@ -104,6 +105,9 @@ class LoRAModel(AdapterModel):
         """Get LoRA for a given module by name"""
         return self.loras.get(module_name, None)
 
+    def check_lora_name(self, lora_name: str) -> bool:
+        return lora_name in self.loras
+
     # (yard1): TODO see if we can derive target_embedding_padding automatically
     @classmethod
     def from_lora_tensors(
@@ -335,6 +339,7 @@ class LoRAModelManager(AdapterModelManager):
         # Used for long context lora.
         self.scaling_factor_to_offset: Dict[float, int] = {}
         super().__init__(model)
+
         self.supported_lora_modules = get_supported_lora_modules(self.model)
         assert self.supported_lora_modules, "No supported LoRA modules found in"
         f"{self.model.__class__.__name__}."
@@ -350,6 +355,7 @@ class LoRAModelManager(AdapterModelManager):
             # In case the model only supports LoRA for
             # text modules (e.g. ChatGLM)
             and hasattr(self.model, "get_mm_mapping"))
+        self.is_pooling_model = is_pooling_model(self.model)
         self.packed_modules: Dict[str, List[str]] = {}
         self.modules: Dict[str, BaseLayerWithLoRA] = {}
         # Dict instead of a Set for compatibility with LRUCache.
@@ -389,7 +395,7 @@ class LoRAModelManager(AdapterModelManager):
                      lora_model.id, index)
         self.lora_index_to_id[index] = lora_model.id
         for module_name, module in self.modules.items():
-            module_lora = lora_model.get_lora(module_name)
+            module_lora = self._get_lora_layer_weights(lora_model, module_name)
             if module_lora:
                 module_lora.optimize()
                 # Bias is not explicitly enabled with the flag enable_lora_bias.
@@ -626,7 +632,7 @@ class LoRAModelManager(AdapterModelManager):
             replaced_module: Set[str] = set()
             has_replacement = False
             for r in new_module_names:
-                lora = lora_model.get_lora(r)
+                lora = self._get_lora_layer_weights(lora_model, r)
                 replacement_loras.append(lora)
                 if lora:
                     has_replacement = True
@@ -637,12 +643,34 @@ class LoRAModelManager(AdapterModelManager):
                 if replacement_loras[i]:
                     continue
                 replacement_loras[i] = None
+            # HACK Temporary solution for the pool model.
+            if self.is_pooling_model and not lora_model.check_lora_name(
+                    module_name):
+                replaced_module_name = module_name.replace("model.", "")
+                if lora_model.check_lora_name(module_name):
+                    module_name = replaced_module_name
             lora_model.loras[module_name] = PackedLoRALayerWeights.pack(
                 replacement_loras)
             # Remove the modules that have been replaced.
             for module in replaced_module:
                 lora_model.loras.pop(module, None)
 
+    def _get_lora_layer_weights(
+            self, lora_model: LoRAModel,
+            module_name: str) -> Optional[LoRALayerWeights]:
+        org_module_name = module_name
+        if self.is_pooling_model and not lora_model.check_lora_name(
+                module_name):
+            # If it's a pool model, and the layer name is not found,
+            # remove the prefix 'model.' and search again.
+            module_name = module_name.replace("model.", "")
+            if lora_model.check_lora_name(module_name):
+                org_module_name = module_name
+                logger.info_once(
+                    "For the pool model, successfully loaded the LoRA weights "
+                    "after removing the prefix 'model.'.")
+        return lora_model.get_lora(org_module_name)
+
     def deactivate_adapter(self, adapter_id: int) -> bool:
         return deactivate_adapter(adapter_id, self._active_adapters,
                                   self._deactivate_adapter)

vllm/model_executor/layers/fused_moe/configs/E=256,N=1024,device_name=AMD_Instinct_MI325X,block_shape=[128,128].json

@@ -0,0 +1,200 @@
+{
+    "1": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 256,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "2": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "4": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "8": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "16": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "24": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "32": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 256,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "48": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 256,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "64": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "96": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 1,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "128": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 1,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "256": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "512": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "1024": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "1536": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "2048": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    }
+}

vllm/model_executor/layers/fused_moe/configs/E=256,N=1024,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,200 @@
+{
+    "1": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 256,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "2": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "4": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "8": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "16": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "24": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "32": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 256,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "48": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 256,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "64": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "96": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 1,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "128": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 1,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "256": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 32,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "512": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "1024": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "1536": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "2048": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 16,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    }
+}

vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,200 @@
+{
+    "1": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 16,
+        "BLOCK_SIZE_K": 256,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "2": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "4": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "8": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "16": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "24": {
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "32": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "48": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 2,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "64": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 2,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "96": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 32,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 4,
+        "num_warps": 2,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "128": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 32,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 4,
+        "num_warps": 1,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "256": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 32,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 2,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "512": {
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "1024": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "1536": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "2048": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    }
+}

vllm/model_executor/layers/fused_moe/configs/E=256,N=256,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,200 @@
+{
+    "1": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 16,
+        "BLOCK_SIZE_K": 256,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "2": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "4": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "8": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "16": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "24": {
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "32": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "48": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 2,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "64": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 2,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "96": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 32,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 4,
+        "num_warps": 2,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "128": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 32,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 4,
+        "num_warps": 1,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "256": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 32,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 2,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "512": {
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "1024": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "1536": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "2048": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    }
+}

vllm/model_executor/layers/fused_moe/configs/E=256,N=512,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,200 @@
+{
+    "1": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 16,
+        "BLOCK_SIZE_K": 256,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "2": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "4": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "8": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "16": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 1
+    },
+    "24": {
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "32": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 4,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "48": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 2,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "64": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 2,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "96": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 32,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 4,
+        "num_warps": 2,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "128": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 32,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 4,
+        "num_warps": 1,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "256": {
+        "BLOCK_SIZE_M": 16,
+        "BLOCK_SIZE_N": 32,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 1,
+        "num_warps": 2,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "512": {
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "1024": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "1536": {
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 128,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "2048": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "3072": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    },
+    "4096": {
+        "BLOCK_SIZE_M": 128,
+        "BLOCK_SIZE_N": 128,
+        "BLOCK_SIZE_K": 64,
+        "GROUP_SIZE_M": 8,
+        "num_warps": 8,
+        "num_stages": 2,
+        "waves_per_eu": 0,
+        "matrix_instr_nonkdim": 16,
+        "kpack": 2
+    }
+}

vllm/model_executor/layers/fused_moe/fused_moe.py

@@ -783,8 +783,12 @@ def invoke_fused_moe_kernel(A: torch.Tensor,
             use_int8_w8a16=use_int8_w8a16,
             **config,
         )
-
     else:
+        config = config.copy()
+        BLOCK_SIZE_K = config.pop("BLOCK_SIZE_K")
+        if block_shape is not None:
+            BLOCK_SIZE_K = min(BLOCK_SIZE_K, min(block_shape[0],
+                                                 block_shape[1]))
         fused_moe_kernel[grid](
             A,
             B,
@@ -823,6 +827,7 @@ def invoke_fused_moe_kernel(A: torch.Tensor,
             compute_type=compute_type,
             use_fp8_w8a8=use_fp8_w8a8,
             use_int8_w8a16=use_int8_w8a16,
+            BLOCK_SIZE_K=BLOCK_SIZE_K,
             **config,
         )
 

vllm/model_executor/layers/quantization/utils/configs/N=1536,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=1536,K=1536,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=1536,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=2048,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=2048,K=512,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=2304,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=2304,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=24576,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=256,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=256,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=3072,K=1536,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=3072,K=1536,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=3072,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=3072,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=32768,K=512,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=36864,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=36864,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=4096,K=512,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=4096,K=512,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=4608,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=4608,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=512,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=512,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=576,K=7168,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1024,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1024,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1152,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=7168,K=1152,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=7168,K=128,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=7168,K=128,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=7168,K=16384,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=7168,K=18432,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2048,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2048,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2304,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=7168,K=2304,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=7168,K=256,device_name=AMD_Instinct_MI325X,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}

vllm/model_executor/layers/quantization/utils/configs/N=7168,K=256,device_name=AMD_Instinct_MI325_OAM,dtype=fp8_w8a8,block_shape=[128,128].json

@@ -0,0 +1,164 @@
+{
+    "1": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "8": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "16": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "24": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "32": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "48": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "64": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 32,
+        "GROUP_SIZE_M": 8,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "96": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "128": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "256": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 64,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "512": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 1,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1024": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "1536": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 32,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 16,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "2048": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "3072": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    },
+    "4096": {
+        "BLOCK_SIZE_K": 128,
+        "BLOCK_SIZE_M": 64,
+        "BLOCK_SIZE_N": 128,
+        "GROUP_SIZE_M": 32,
+        "kpack": 1,
+        "matrix_instr_nonkdim": 16,
+        "num_warps": 4
+    }
+}