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

[Model] Enable optional prefix when loading embedding models (#10639)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2024-11-26 02:14:33 +08:00
committed by GitHub
parent b1d920531f
commit cf73f0c95e
5 changed files with 20 additions and 13 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
vllm/model_executor/models/qwen2.py
View File

@@ -50,7 +50,8 @@ from vllm.model_executor.sampling_metadata import SamplingMetadata
from vllm.sequence import IntermediateTensors, PoolerOutput
from .interfaces import SupportsLoRA, SupportsPP
from .utils import (AutoWeightsLoader, PPMissingLayer, is_pp_missing_parameter,
from .utils import (AutoWeightsLoader, PPMissingLayer, WeightsMapper,
is_pp_missing_parameter,
make_empty_intermediate_tensors_factory, make_layers,
maybe_prefix)
@@ -585,8 +586,7 @@ class Qwen2EmbeddingModel(nn.Module, SupportsLoRA, SupportsPP):
) -> Optional[PoolerOutput]:
return self._pooler(hidden_states, pooling_metadata)
def load_weights(self, weights: Iterable[Tuple[str,
torch.Tensor]]) -> Set[str]:
loader = AutoWeightsLoader(self,
ignore_unexpected_prefixes=["lm_head."])
return loader.load_weights(weights)
def load_weights(self, weights: Iterable[Tuple[str, torch.Tensor]]):
hf_to_vllm_mapper = WeightsMapper(orig_to_new_prefix={"model.": ""})
weights = hf_to_vllm_mapper.apply(weights)
self.model.load_weights(weights)