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

[PERF] PyTorch Symmetric Memory All-Reduce (#20759)

Browse Source 
Signed-off-by: ilmarkov <imarkov@redhat.com>
Signed-off-by: ilmarkov <markovilya197@gmail.com>
Signed-off-by: Michael Goin <mgoin64@gmail.com>
Co-authored-by: ilmarkov <imarkov@redhat.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>
This commit is contained in:
Ilya Markov
2025-08-22 23:39:08 +02:00
committed by GitHub
parent 0483fabc74
commit 0313cf854d
8 changed files with 283 additions and 5 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
vllm/envs.py
View File

@@ -161,6 +161,7 @@ if TYPE_CHECKING:
VLLM_HAS_FLASHINFER_CUBIN: bool = False
VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8: bool = False
VLLM_USE_FLASHINFER_MOE_MXFP4_BF16: bool = False
VLLM_ALLREDUCE_USE_SYMM_MEM: bool = False
VLLM_TUNED_CONFIG_FOLDER: Optional[str] = None
@@ -1156,6 +1157,10 @@ environment_variables: dict[str, Callable[[], Any]] = {
"VLLM_ENABLE_RESPONSES_API_STORE":
lambda: bool(int(os.getenv("VLLM_ENABLE_RESPONSES_API_STORE", "0"))),
# Whether to use pytorch symmetric memory for allreduce
"VLLM_ALLREDUCE_USE_SYMM_MEM":
lambda: bool(int(os.getenv("VLLM_ALLREDUCE_USE_SYMM_MEM", "0"))),
# Allows vllm to find tuned config under customized folder
"VLLM_TUNED_CONFIG_FOLDER":
lambda: os.getenv("VLLM_TUNED_CONFIG_FOLDER", None),