xinyun/vllm
1
0
Fork 0
mirror of https://git.datalinker.icu/vllm-project/vllm.git synced 2025-12-10 02:05:01 +08:00
Code Issues Packages Projects Releases Wiki Activity

[Bugfix] Lower gpt-oss max cudagraph size to 992 to be compatible with FA3 (#25508)

Browse Source 
Signed-off-by: mgoin <mgoin64@gmail.com>
This commit is contained in:
Michael Goin 2025-09-23 15:49:55 -04:00 committed by GitHub
parent d5944d5146
commit a8ffc4f0f2
No known key found for this signature in database
GPG Key ID: B5690EEEBB952194
1 changed files with 5 additions and 5 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

10
vllm/model_executor/models/config.py
View File

@ -266,24 +266,24 @@ class GptOssForCausalLMConfig(VerifyAndUpdateConfig):
if structured_outputs_config.reasoning_parser == "":
structured_outputs_config.reasoning_parser = "openai_gptoss"
# Increase the max capture size from 512 to 1024 for performance.
# Increase the max capture size from 512 to 992 for performance.
# NOTE(woosuk): This will increase the number of CUDA graphs
# from 67 to 83.
# from 67 to 81.
scheduler_config = vllm_config.scheduler_config
if len(scheduler_config.cuda_graph_sizes) == 1:
max_capture_size = scheduler_config.cuda_graph_sizes[0]
# FIXME(woosuk): When using full cuda graph with FA3, the max
# supported size is 992.
if max_capture_size < 1024:
if max_capture_size < 992:
cuda_graph_sizes = [1, 2, 4]
# Step size 8 for small batch sizes
cuda_graph_sizes += [i for i in range(8, 256, 8)]
# Step size 16 for larger batch sizes
cuda_graph_sizes += [i for i in range(256, 1025, 16)]
cuda_graph_sizes += [i for i in range(256, 993, 16)]
scheduler_config.cuda_graph_sizes = cuda_graph_sizes
logger.info(
"Overriding max cuda graph capture size to "
"%d for performance.", 1024)
"%d for performance.", 992)
class MambaModelConfig(VerifyAndUpdateConfig):