A high-throughput and memory-efficient inference and serving engine for LLMs
Mirrored from https://github.com/tduncan-lum/Nvidia-Optimization-Stack---VLLM.git
Synced Nov 14, 05:30 UTC
[Log] Add Warning for Deprecation of DeepGEMM old version (#22194)
Signed-off-by: yewentao256 <zhyanwentao@126.com>
Wentao Ye committed
Aug 9, 2025 at 02:07 UTC
3157aebb63a2e121da6de943754dc95dffd14caa
1 file changed
+9
vllm/utils/deep_gemm.py
+9
@@ -14,6 +14,7 @@ from typing import Any, Callable, NoReturn
14
import torch
15
16
import vllm.envs as envs
17
+from vllm.logger import logger
18
from vllm.platforms import current_platform
19
from vllm.utils import cdiv, has_deep_gemm
20
@@ -57,6 +58,14 @@ def _resolve_symbol(module, new: str, old: str) -> Callable[..., Any] | None:
58
if hasattr(module, new):
59
return getattr(module, new)
60
if hasattr(module, old):
61
+ # TODO(wentao): deprecate old symbol in the future.
62
+ logger.warning_once(
63
+ "Found legacy DeepGEMM symbol `%s`. Please upgrade the `deep_gemm` "
64
+ "package so that `%s` is available. Support for the legacy symbol "
65
+ "will be removed in a future vLLM release.",
66
+ old,
67
+ new,
68
+ )
69
return getattr(module, old)
70
return None
71