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

[Model][2/N] Improve all pooling task | Support multi-vector retrieval (#25370)

Browse Source 
Signed-off-by: wang.yuqi <noooop@126.com>
This commit is contained in:
wang.yuqi
2025-10-15 19:14:41 +08:00
committed by GitHub
parent d4d1a6024f
commit f54f85129e
41 changed files with 786 additions and 399 deletions
Download Patch File Download Diff File Expand all files Collapse all files

26
vllm/model_executor/models/roberta.py
View File

@@ -105,15 +105,7 @@ class RobertaClassificationHead(nn.Module):
@default_pooling_type("CLS")
class RobertaEmbeddingModel(BertEmbeddingModel):
"""A model that uses Roberta to provide embedding functionalities.
This class encapsulates the BertModel and provides an interface for
embedding operations and customized pooling functions.
Attributes:
model: An instance of BertModel used for forward operations.
_pooler: An instance of Pooler used for pooling operations.
"""
"""A model that uses Roberta to provide embedding functionalities."""
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__(vllm_config=vllm_config, prefix=prefix)
@@ -212,20 +204,14 @@ class RobertaForSequenceClassification(nn.Module, SupportsCrossEncoding):
self.pooler = DispatchPooler(
{
"encode": Pooler.for_encode(pooler_config),
"token_classify": Pooler.for_token_classify(
pooler_config=pooler_config, classifier=self.classifier
),
"classify": ClassifierPooler(
pooling=CLSPool(),
classifier=self.classifier,
act_fn=ClassifierPooler.act_fn_for_seq_cls(
vllm_config.model_config
),
pooling=CLSPool(), classifier=self.classifier, act_fn="classify"
),
"score": ClassifierPooler(
pooling=CLSPool(),
classifier=self.classifier,
act_fn=ClassifierPooler.act_fn_for_cross_encoder(
vllm_config.model_config
),
pooling=CLSPool(), classifier=self.classifier, act_fn="score"
),
}
)