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

8
vllm/config/__init__.py
View File

@@ -871,6 +871,10 @@ class ModelConfig:
if getattr(pooler_config, k) is None:
setattr(pooler_config, k, v)
default_pooling_type = self._model_info.default_pooling_type
if pooler_config.pooling_type is None:
pooler_config.pooling_type = default_pooling_type
return pooler_config
return None
@@ -3844,6 +3848,10 @@ class VllmConfig:
disable_chunked_prefill_reasons.append(
"Only \"last\" pooling supports chunked "
"prefill and prefix caching; disabling both.")
elif not getattr(self.model_config.hf_config, "is_causal", True):
disable_chunked_prefill_reasons.append(
"Only models using causal attention supports chunked "
"prefill and prefix caching; disabling both.")
if disable_chunked_prefill_reasons:
for reason in disable_chunked_prefill_reasons: