chenchenghao/sglang
1
0
Fork 0
Code Issues Pull Requests Actions 12 Packages Projects Releases Wiki Activity

vlm: enable radix cache for qwen-vl models (#5349)

Browse Source 
Co-authored-by: Xinyuan Tong <justinning0323@outlook.com>
This commit is contained in:
Mick
2025-04-24 12:35:05 +09:00
committed by GitHub
parent 7d0edf3cae
commit c998d04b46
26 changed files with 429 additions and 331 deletions
Download Patch File Download Diff File Expand all files Collapse all files

9
python/sglang/srt/model_executor/model_runner.py
View File

@@ -310,15 +310,6 @@ class ModelRunner:
)
server_args.chunked_prefill_size = -1
if self.model_config.hf_config.architectures == [
"Qwen2VLForConditionalGeneration"
] or self.model_config.hf_config.architectures == [
"Qwen2_5_VLForConditionalGeneration"
]:
# TODO: qwen2-vl series does not support radix cache now, set disable_radix_cache=True automatically
logger.info("Automatically disable radix cache for qwen-vl series.")
server_args.disable_radix_cache = True
if server_args.enable_deepep_moe:
logger.info(f"DeepEP is turned on. DeepEP mode: {server_args.deepep_mode}")