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

[Bugfix] Fix Qwen2-VL LoRA weight loading (#11430)

Browse Source 
Signed-off-by: Jee Jee Li <pandaleefree@gmail.com>
This commit is contained in:
Jee Jee Li
2024-12-24 17:56:10 +08:00
committed by GitHub
parent 9edca6bf8f
commit b1b1038fbd
7 changed files with 168 additions and 14 deletions
Download Patch File Download Diff File Expand all files Collapse all files

12
vllm/model_executor/models/qwen2_vl.py
View File

@@ -901,6 +901,11 @@ class Qwen2VLForConditionalGeneration(nn.Module, SupportsMultiModal,
]
embedding_modules = {}
embedding_padding_modules = []
# To ensure correct weight loading and mapping.
hf_to_vllm_mapper = WeightsMapper(orig_to_new_prefix={
"lm_head.": "language_model.lm_head.",
"model.": "language_model.model.",
})
def __init__(self, *, vllm_config: VllmConfig, prefix: str = ""):
super().__init__()
@@ -1190,11 +1195,6 @@ class Qwen2VLForConditionalGeneration(nn.Module, SupportsMultiModal,
def load_weights(self, weights: Iterable[Tuple[str,
torch.Tensor]]) -> Set[str]:
hf_to_vllm_mapper = WeightsMapper(
orig_to_new_prefix={
"lm_head.": "language_model.lm_head.",
"model.": "language_model.model.",
})
loader = AutoWeightsLoader(self)
return loader.load_weights(weights, mapper=hf_to_vllm_mapper)
return loader.load_weights(weights, mapper=self.hf_to_vllm_mapper)