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
Remove redundant cudagraph dispatcher warning (#25841)
Michael Goin committed
Sep 28, 2025 at 17:12 UTC
0307428d65acf5cf1a73a70a7722e076bbb83f22
1 file changed
-5
vllm/v1/cudagraph_dispatcher.py
-5
@@ -4,9 +4,6 @@ from typing import Optional
4
5
from vllm.config import CUDAGraphMode, VllmConfig
6
from vllm.forward_context import BatchDescriptor
7
-from vllm.logger import init_logger
8
-
9
-logger = init_logger(__name__)
7
8
9
class CudagraphDispatcher:
@@ -102,8 +99,6 @@ class CudagraphDispatcher:
99
"""
100
# if not initialized, just skip dispatching.
101
if not self.keys_initialized:
105
- logger.warning_once("cudagraph dispatching keys are not "
106
- "initialized. No cudagraph will be used.")
102
return CUDAGraphMode.NONE, None
103
104
non_uniform_key = batch_descriptor.non_uniform