xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-20 04:05:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Refactor] Remove VLLM_DEEPEP_LOW_LATENCY_ALLOW_NVLINK (#27750)

Browse Source 
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye 2025-10-30 15:32:39 -04:00 committed by GitHub
parent 4917002523
commit a8141fa649
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
2 changed files with 1 additions and 8 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

2
vllm/distributed/device_communicators/all2all.py
View File

@ -363,7 +363,7 @@ class DeepEPLLAll2AllManager(DeepEPAll2AllManagerBase):
num_rdma_bytes=num_rdma_bytes, num_rdma_bytes=num_rdma_bytes,
low_latency_mode=True, low_latency_mode=True,
num_qps_per_rank=num_qps_per_rank, num_qps_per_rank=num_qps_per_rank,
allow_nvlink_for_low_latency_mode=envs.VLLM_DEEPEP_LOW_LATENCY_ALLOW_NVLINK, allow_nvlink_for_low_latency_mode=True,
allow_mnnvl=envs.VLLM_DEEPEP_LOW_LATENCY_USE_MNNVL, allow_mnnvl=envs.VLLM_DEEPEP_LOW_LATENCY_USE_MNNVL,
) )

7
vllm/envs.py
View File

@ -207,7 +207,6 @@ if TYPE_CHECKING:
VLLM_OBJECT_STORAGE_SHM_BUFFER_NAME: str = "VLLM_OBJECT_STORAGE_SHM_BUFFER" VLLM_OBJECT_STORAGE_SHM_BUFFER_NAME: str = "VLLM_OBJECT_STORAGE_SHM_BUFFER"
VLLM_DEEPEP_BUFFER_SIZE_MB: int = 1024 VLLM_DEEPEP_BUFFER_SIZE_MB: int = 1024
VLLM_DEEPEP_HIGH_THROUGHPUT_FORCE_INTRA_NODE: bool = False VLLM_DEEPEP_HIGH_THROUGHPUT_FORCE_INTRA_NODE: bool = False
VLLM_DEEPEP_LOW_LATENCY_ALLOW_NVLINK: bool = True
VLLM_DEEPEP_LOW_LATENCY_USE_MNNVL: bool = False VLLM_DEEPEP_LOW_LATENCY_USE_MNNVL: bool = False
VLLM_DBO_COMM_SMS: int = 20 VLLM_DBO_COMM_SMS: int = 20
VLLM_PATTERN_MATCH_DEBUG: str | None = None VLLM_PATTERN_MATCH_DEBUG: str | None = None
@ -1400,11 +1399,6 @@ environment_variables: dict[str, Callable[[], Any]] = {
"VLLM_DEEPEP_HIGH_THROUGHPUT_FORCE_INTRA_NODE": lambda: bool( "VLLM_DEEPEP_HIGH_THROUGHPUT_FORCE_INTRA_NODE": lambda: bool(
int(os.getenv("VLLM_DEEPEP_HIGH_THROUGHPUT_FORCE_INTRA_NODE", "0")) int(os.getenv("VLLM_DEEPEP_HIGH_THROUGHPUT_FORCE_INTRA_NODE", "0"))
), ),
# Allow DeepEP to use nvlink for internode_ll kernel, turn this on for
# better latency on GB200 like system
"VLLM_DEEPEP_LOW_LATENCY_ALLOW_NVLINK": lambda: bool(
int(os.getenv("VLLM_DEEPEP_LOW_LATENCY_ALLOW_NVLINK", "1"))
),
# Allow DeepEP to use MNNVL (multi-node nvlink) for internode_ll kernel, # Allow DeepEP to use MNNVL (multi-node nvlink) for internode_ll kernel,
# turn this for better latency on GB200 like system # turn this for better latency on GB200 like system
"VLLM_DEEPEP_LOW_LATENCY_USE_MNNVL": lambda: bool( "VLLM_DEEPEP_LOW_LATENCY_USE_MNNVL": lambda: bool(
@ -1566,7 +1560,6 @@ def compute_hash() -> str:
"VLLM_NVFP4_GEMM_BACKEND", "VLLM_NVFP4_GEMM_BACKEND",
"VLLM_USE_FBGEMM", "VLLM_USE_FBGEMM",
"VLLM_DEEPEP_HIGH_THROUGHPUT_FORCE_INTRA_NODE", "VLLM_DEEPEP_HIGH_THROUGHPUT_FORCE_INTRA_NODE",
"VLLM_DEEPEP_LOW_LATENCY_ALLOW_NVLINK",
"VLLM_DEEPEP_LOW_LATENCY_USE_MNNVL", "VLLM_DEEPEP_LOW_LATENCY_USE_MNNVL",
] ]
for key in environment_variables_to_hash: for key in environment_variables_to_hash: