[BugFix] Graceful handling of torch symm mem errors. (#27671)
Signed-off-by: ilmarkov <markovilya197@gmail.com> Co-authored-by: Michael Goin <mgoin64@gmail.com>
This commit is contained in:
co-authored by
Michael Goin
parent
d23539549a
commit
1788aa1efb
+2
-2
@@ -201,7 +201,7 @@ if TYPE_CHECKING:
|
||||
VLLM_USE_FLASHINFER_MOE_MXFP4_BF16: bool = False
|
||||
VLLM_ROCM_FP8_MFMA_PAGE_ATTN: bool = False
|
||||
VLLM_USE_FLASHINFER_MOE_MXFP4_MXFP8_CUTLASS: bool = False
|
||||
VLLM_ALLREDUCE_USE_SYMM_MEM: bool = False
|
||||
VLLM_ALLREDUCE_USE_SYMM_MEM: bool = True
|
||||
VLLM_TUNED_CONFIG_FOLDER: str | None = None
|
||||
VLLM_GPT_OSS_SYSTEM_TOOL_MCP_LABELS: set[str] = set()
|
||||
VLLM_GPT_OSS_HARMONY_SYSTEM_INSTRUCTIONS: bool = False
|
||||
@@ -1389,7 +1389,7 @@ environment_variables: dict[str, Callable[[], Any]] = {
|
||||
),
|
||||
# Whether to use pytorch symmetric memory for allreduce
|
||||
"VLLM_ALLREDUCE_USE_SYMM_MEM": lambda: bool(
|
||||
int(os.getenv("VLLM_ALLREDUCE_USE_SYMM_MEM", "0"))
|
||||
int(os.getenv("VLLM_ALLREDUCE_USE_SYMM_MEM", "1"))
|
||||
),
|
||||
# Allows vllm to find tuned config under customized folder
|
||||
"VLLM_TUNED_CONFIG_FOLDER": lambda: os.getenv("VLLM_TUNED_CONFIG_FOLDER", None),
|
||||
|
||||
Reference in New Issue
Block a user