biondizzle/vllm
1
0
Fork 0
Code Issues Pull Requests Actions 2 Packages Projects Releases Wiki Activity

[Model] Pooling models default to using chunked prefill & prefix caching if supported. (#20930)

Browse Source 
Signed-off-by: wang.yuqi <noooop@126.com>
This commit is contained in:
wang.yuqi
2025-08-12 00:41:37 +08:00
committed by GitHub
parent 16fb668b61
commit 84cf78acee
31 changed files with 452 additions and 261 deletions
Download Patch File Download Diff File Expand all files Collapse all files

4
vllm/model_executor/models/adapters.py
View File

@@ -182,8 +182,8 @@ def as_seq_cls_model(cls: _T) -> _T:
assert pooler_config is not None
pooling_type_str = pooler_config.pooling_type
pooling_type = (PoolingType.LAST if pooling_type_str is None else
PoolingType[pooling_type_str])
assert pooling_type_str is not None
pooling_type = PoolingType[pooling_type_str]
self.pooler = DispatchPooler({
"encode":