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

Enable headless models for pooling in the Transformers backend (#21767)

Browse Source 
Signed-off-by: Harry Mellor <19981378+hmellor@users.noreply.github.com>
This commit is contained in:
Harry Mellor
2025-08-01 18:31:29 +01:00
committed by GitHub
parent ac45c44d98
commit 38c8bce8b6
5 changed files with 44 additions and 9 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/model_executor/models/registry.py
View File

@@ -270,8 +270,9 @@ _TRANSFORMERS_SUPPORTED_MODELS = {
}
_TRANSFORMERS_BACKEND_MODELS = {
"TransformersForMultimodalLM": ("transformers", "TransformersForMultimodalLM"), # noqa: E501
"TransformersModel": ("transformers", "TransformersModel"),
"TransformersForCausalLM": ("transformers", "TransformersForCausalLM"),
"TransformersForMultimodalLM": ("transformers", "TransformersForMultimodalLM"), # noqa: E501
}
# yapf: enable

12
vllm/model_executor/models/transformers.py
View File

@@ -651,6 +651,18 @@ class TransformersBase(nn.Module, SupportsQuant, SupportsLoRA, SupportsPP):
return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)
@support_torch_compile
class TransformersModel(TransformersBase):
hf_to_vllm_mapper = WeightsMapper(
orig_to_new_prefix={
# Add `model.` prefix for base model checkpoints
"": "model.",
# Remove `model.` from places it should not be
"model.model.": "model.",
"model.score": "score",
})
@support_torch_compile
class TransformersForCausalLM(TransformersBase):