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

8
vllm/envs.py
View File

@@ -34,6 +34,7 @@ if TYPE_CHECKING:
VLLM_OPENVINO_ENABLE_QUANTIZED_WEIGHTS: bool = False
VLLM_XLA_CACHE_PATH: str = os.path.join(VLLM_CACHE_ROOT, "xla_cache")
VLLM_FUSED_MOE_CHUNK_SIZE: int = 64 * 1024
VLLM_USE_RAY_SPMD_WORKER: bool = False
VLLM_USE_RAY_COMPILED_DAG: bool = False
VLLM_WORKER_MULTIPROC_METHOD: str = "fork"
VLLM_ASSETS_CACHE: str = os.path.join(VLLM_CACHE_ROOT, "assets")
@@ -261,6 +262,13 @@ environment_variables: Dict[str, Callable[[], Any]] = {
"VLLM_OPENVINO_ENABLE_QUANTIZED_WEIGHTS":
lambda: bool(os.getenv("VLLM_OPENVINO_ENABLE_QUANTIZED_WEIGHTS", False)),
# If the env var is set, then all workers will execute as separate
# processes from the engine, and we use the same mechanism to trigger
# execution on all workers.
# Run vLLM with VLLM_USE_RAY_SPMD_WORKER=1 to enable it.
"VLLM_USE_RAY_SPMD_WORKER":
lambda: bool(os.getenv("VLLM_USE_RAY_SPMD_WORKER", 0)),
# If the env var is set, it uses the Ray's compiled DAG API
# which optimizes the control plane overhead.
# Run vLLM with VLLM_USE_RAY_COMPILED_DAG=1 to enable it.