biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Core] Introduce SPMD worker execution using Ray accelerated DAG (#6032)

Browse Source 
Signed-off-by: Rui Qiao <ruisearch42@gmail.com>
Co-authored-by: Stephanie Wang <swang@cs.berkeley.edu>
This commit is contained in:
Rui Qiao
2024-07-17 22:27:09 -07:00
committed by GitHub
parent d25877dd9b
commit 61e592747c
8 changed files with 216 additions and 119 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
.buildkite/test-pipeline.yaml
View File

@@ -84,6 +84,8 @@ steps:
- VLLM_TEST_SAME_HOST=1 torchrun --nproc-per-node=4 distributed/test_same_node.py
- TEST_DIST_MODEL=facebook/opt-125m DISTRIBUTED_EXECUTOR_BACKEND=ray pytest -v -s distributed/test_basic_distributed_correctness.py
- TEST_DIST_MODEL=meta-llama/Llama-2-7b-hf DISTRIBUTED_EXECUTOR_BACKEND=ray pytest -v -s distributed/test_basic_distributed_correctness.py
- TEST_DIST_MODEL=facebook/opt-125m DISTRIBUTED_EXECUTOR_BACKEND=ray VLLM_USE_RAY_SPMD_WORKER=1 VLLM_USE_RAY_COMPILED_DAG=1 pytest -v -s distributed/test_basic_distributed_correctness.py
- TEST_DIST_MODEL=meta-llama/Llama-2-7b-hf DISTRIBUTED_EXECUTOR_BACKEND=ray VLLM_USE_RAY_SPMD_WORKER=1 VLLM_USE_RAY_COMPILED_DAG=1 pytest -v -s distributed/test_basic_distributed_correctness.py
- TEST_DIST_MODEL=facebook/opt-125m DISTRIBUTED_EXECUTOR_BACKEND=ray pytest -v -s distributed/test_chunked_prefill_distributed.py
- TEST_DIST_MODEL=meta-llama/Llama-2-7b-hf DISTRIBUTED_EXECUTOR_BACKEND=ray pytest -v -s distributed/test_chunked_prefill_distributed.py
- TEST_DIST_MODEL=llava-hf/llava-1.5-7b-hf DISTRIBUTED_EXECUTOR_BACKEND=ray pytest -v -s distributed/test_multimodal_broadcast.py
@@ -108,6 +110,7 @@ steps:
# We want to test that models which use 2 GPUs work with 4 GPUs, which is why we duplicate them here.
# See https://github.com/vllm-project/vllm/pull/5473#issuecomment-2166601837 for context.
- TEST_DIST_MODEL=facebook/opt-125m DISTRIBUTED_EXECUTOR_BACKEND=ray pytest -v -s distributed/test_basic_distributed_correctness.py
- TEST_DIST_MODEL=facebook/opt-125m DISTRIBUTED_EXECUTOR_BACKEND=ray VLLM_USE_RAY_SPMD_WORKER=1 VLLM_USE_RAY_COMPILED_DAG=1 pytest -v -s distributed/test_basic_distributed_correctness.py
- TEST_DIST_MODEL=facebook/opt-125m DISTRIBUTED_EXECUTOR_BACKEND=mp pytest -v -s distributed/test_basic_distributed_correctness.py
- pytest -v -s spec_decode/e2e/test_integration_dist_tp4.py