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

[Core] More fixes to MultiModalEmbeddings type handling (#19715)

Browse Source 
Signed-off-by: Russell Bryant <rbryant@redhat.com>
This commit is contained in:
Russell Bryant
2025-06-18 18:48:29 -04:00
committed by GitHub
parent 04fefe7c9a
commit 14fdd21d39
35 changed files with 71 additions and 36 deletions
Download Patch File Download Diff File Expand all files Collapse all files

3
vllm/model_executor/models/blip2.py
View File

@@ -641,7 +641,8 @@ class Blip2ForConditionalGeneration(nn.Module, SupportsMultiModal, SupportsPP,
multimodal_embeddings: Optional[MultiModalEmbeddings] = None,
) -> torch.Tensor:
inputs_embeds = self.language_model.get_input_embeddings(input_ids)
if multimodal_embeddings is not None:
if multimodal_embeddings is not None \
and len(multimodal_embeddings) != 0:
inputs_embeds = merge_multimodal_embeddings(
input_ids, inputs_embeds, multimodal_embeddings,
_IMAGE_TOKEN_ID)