Support embedding models in V1 (#16188)

Signed-off-by: Max de Bayser <mbayser@br.ibm.com>
Signed-off-by: Max de Bayser <maxdebayser@gmail.com>
Signed-off-by: 22quinn <33176974+22quinn@users.noreply.github.com>
Co-authored-by: 22quinn <33176974+22quinn@users.noreply.github.com>
This commit is contained in:
Maximilien de Bayser
2025-06-18 21:36:33 -07:00
committed by GitHub
co-authored by 22quinn
parent 4959915089
commit 799397ee4f
56 changed files with 889 additions and 281 deletions
+2 -2
View File
@@ -446,8 +446,8 @@ class BertEmbeddingModel(nn.Module, SupportsV0Only, SupportsQuant):
softmax=False)
class BertForSequenceClassification(nn.Module, SupportsCrossEncoding,
SupportsQuant):
class BertForSequenceClassification(nn.Module, SupportsV0Only,
SupportsCrossEncoding, SupportsQuant):
"""A model that uses Bert to provide embedding functionalities.
This class encapsulates the BertModel and provides an interface for
+3 -2
View File
@@ -21,7 +21,7 @@ from vllm.model_executor.model_loader.weight_utils import default_weight_loader
from vllm.model_executor.pooling_metadata import PoolingMetadata
from vllm.sequence import IntermediateTensors, PoolerOutput
from .interfaces import SupportsCrossEncoding
from .interfaces import SupportsCrossEncoding, SupportsV0Only
from .utils import WeightsMapper, maybe_prefix
@@ -270,7 +270,8 @@ class ModernBertPooler(nn.Module):
return pooled_output
class ModernBertForSequenceClassification(nn.Module, SupportsCrossEncoding):
class ModernBertForSequenceClassification(nn.Module, SupportsV0Only,
SupportsCrossEncoding):
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__()
+6 -1
View File
@@ -375,7 +375,12 @@ class Qwen3ForSequenceClassification(nn.Module, SupportsLoRA,
) -> Optional[PoolerOutput]:
hidden_states = self._pooler.extract_states(hidden_states,
pooling_metadata)
logits, _ = self.score(hidden_states)
if isinstance(hidden_states, list):
logits = [self.score(state)[0] for state in hidden_states]
else:
logits, _ = self.score(hidden_states)
pooled_data = self._pooler.head(logits, pooling_metadata)
pooled_outputs = [
self._pooler.build_output(data.squeeze(-1)) for data in pooled_data