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

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
2025-10-14 01:27:01 +08:00
parent e3b90c1ba2
commit afc47e4de7
11 changed files with 127 additions and 331 deletions
--- a/vllm/model_executor/models/skyworkr1v.py
+++ b/vllm/model_executor/models/skyworkr1v.py
@@ -814,8 +814,11 @@ class SkyworkR1VChatModel(nn.Module, SupportsMultiModal, SupportsPP):
            )

        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:
            return SkyworkR1VImagePixelInputs(