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

Move mem_fraction_static adjustment for multimodal models to server_args.py & Fix session control & Other cleanups (#7748)

Browse Source
This commit is contained in:
Lianmin Zheng
2025-07-04 16:33:33 -07:00
committed by GitHub
parent 975a5ec69c
commit 14229ccf8f
16 changed files with 339 additions and 137 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
python/sglang/srt/model_executor/model_runner.py
View File

@@ -451,11 +451,6 @@ class ModelRunner:
self.init_double_sparsity_channel_config(server_args.ds_heavy_channel_type)
if self.is_multimodal:
self.mem_fraction_static *= 0.90
logger.info(
f"Automatically reduce --mem-fraction-static to {self.mem_fraction_static:.3f} "
f"because this is a multimodal model."
)
if not self.is_multimodal_chunked_prefill_supported:
server_args.chunked_prefill_size = -1
logger.info(