chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 11 Packages Projects Releases Wiki Activity

chore: tune mem fraction static for vlm (#6881)

Browse Source
This commit is contained in:
Mick
2025-07-19 08:19:27 +08:00
committed by GitHub
parent 9c7a46180c
commit 3964b352c3
4 changed files with 57 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
python/sglang/srt/model_executor/model_runner.py
View File

@@ -411,7 +411,7 @@ class ModelRunner:
else:
server_args.attention_backend = "triton"
logger.info(
f"Attention backend not set. Use {server_args.attention_backend} backend by default."
f"Attention backend not explicitly specified. Use {server_args.attention_backend} backend by default."
)
elif self.use_mla_backend:
if server_args.device != "cpu":
@@ -463,7 +463,7 @@ class ModelRunner:
if not self.is_multimodal_chunked_prefill_supported:
server_args.chunked_prefill_size = -1
logger.info(
f"Automatically turn of --chunked-prefill-size as it is not supported for "
f"Automatically turn off --chunked-prefill-size as it is not supported for "
f"{self.model_config.hf_config.model_type}"
)