xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2026-01-23 10:34:27 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Deprecation] Remove DeepGEMM Old Symbol Wrapper (#24902)

Browse Source 
Signed-off-by: yewentao256 <zhyanwentao@126.com>
This commit is contained in:
Wentao Ye 2025-09-15 22:03:29 -04:00 committed by GitHub
parent e95084308b
commit de2cc3d867
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 3 additions and 25 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

28
vllm/utils/deep_gemm.py
View File

@ -67,23 +67,6 @@ def _missing(*_: Any, **__: Any) -> NoReturn:
"package to enable FP8 kernels.")
def _resolve_symbol(module, new: str, old: str) -> Callable[..., Any] | None:
"""Return the *new* symbol if it exists, otherwise the *old* one."""
if hasattr(module, new):
return getattr(module, new)
if hasattr(module, old):
# TODO(wentao): deprecate old symbol in the future.
logger.warning_once(
"Found legacy DeepGEMM symbol `%s`. Please upgrade the `deep_gemm` "
"package so that `%s` is available. Support for the legacy symbol "
"will be removed in a future vLLM release.",
old,
new,
)
return getattr(module, old)
return None
_fp8_gemm_nt_impl: Callable[..., Any] | None = None
_grouped_impl: Callable[..., Any] | None = None
_grouped_masked_impl: Callable[..., Any] | None = None
@ -109,14 +92,9 @@ def _lazy_init() -> None:
_dg = importlib.import_module("deep_gemm")
_fp8_gemm_nt_impl = _resolve_symbol(_dg, "fp8_gemm_nt",
"gemm_fp8_fp8_bf16_nt")
_grouped_impl = _resolve_symbol(
_dg, "m_grouped_fp8_gemm_nt_contiguous",
"m_grouped_gemm_fp8_fp8_bf16_nt_contiguous")
_grouped_masked_impl = _resolve_symbol(
_dg, "fp8_m_grouped_gemm_nt_masked",
"m_grouped_gemm_fp8_fp8_bf16_nt_masked")
_fp8_gemm_nt_impl = getattr(_dg, "fp8_gemm_nt", None)
_grouped_impl = getattr(_dg, "m_grouped_fp8_gemm_nt_contiguous", None)
_grouped_masked_impl = getattr(_dg, "fp8_m_grouped_gemm_nt_masked", None)
def fp8_gemm_nt(*args, **kwargs):