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

[Model] Use merge_by_field_config for MM models (M-N) (#26710)

Browse Source 
Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2025-10-14 01:27:01 +08:00
committed by GitHub
parent e3b90c1ba2
commit afc47e4de7
11 changed files with 127 additions and 331 deletions
Download Patch File Download Diff File Expand all files Collapse all files

14
vllm/model_executor/models/internvl.py
View File

@@ -1232,8 +1232,11 @@ class InternVLChatModel(nn.Module, SupportsMultiModal, SupportsPP, SupportsLoRA)
)
image_token_id = kwargs["image_token_id"]
assert isinstance(image_token_id, torch.Tensor)
self.img_context_token_id = image_token_id.flatten().unique().item()
if isinstance(image_token_id, torch.Tensor):
image_token_id = image_token_id.flatten().unique().item()
assert isinstance(image_token_id, int)
self.img_context_token_id = image_token_id
if pixel_values_flat is not None:
expected_h = expected_w = self.config.vision_config.image_size
@@ -1265,8 +1268,11 @@ class InternVLChatModel(nn.Module, SupportsMultiModal, SupportsPP, SupportsLoRA)
)
video_token_id = kwargs["video_token_id"]
assert isinstance(video_token_id, torch.Tensor)
self.video_context_token_id = video_token_id.flatten().unique().item()
if isinstance(video_token_id, torch.Tensor):
video_token_id = video_token_id.flatten().unique().item()
assert isinstance(video_token_id, int)
self.video_context_token_id = video_token_id
if pixel_values_flat_video is not None:
expected_h = expected_w = self.config.vision_config.image_size