[Bugfix] Fix the two multimodal root tests that fail on main (OpenPangu-VL embed merge, MiMo sink test fixture) (#58900)

Signed-off-by: khluu <khluu000@gmail.com>
Co-authored-by: Claude Opus 5.5 <noreply@anthropic.com>
This commit is contained in:
Kevin H. Luu
2026-09-28 04:52:51 +00:00
committed by GitHub
co-authored by Claude Opus 5.5
parent dd3bc9c72f
commit 0da126676e
2 changed files with 15 additions and 27 deletions
+12 -9
View File
@@ -1055,15 +1055,18 @@ class OpenPanguVLForConditionalGeneration(
input_ids: torch.Tensor,
multimodal_embeddings: MultiModalEmbeddings | None = None,
) -> torch.Tensor:
inputs_embeds = self.language_model.embed_input_ids(input_ids)
if multimodal_embeddings is not None:
inputs_embeds = self.embed_input_ids( # type: ignore[call-overload]
input_ids,
inputs_embeds,
multimodal_embeddings,
[self.config.image_token_id, self.config.video_token_id],
)
return inputs_embeds
if multimodal_embeddings is None:
return self.language_model.embed_input_ids(input_ids)
mm_token_ids = input_ids.new_tensor(
[self.config.image_token_id, self.config.video_token_id]
)
is_multimodal = torch.isin(input_ids, mm_token_ids)
return self.embed_input_ids(
input_ids,
multimodal_embeddings,
is_multimodal=is_multimodal,
)
def _process_image_input(
self, image_input: OpenPanguVLImageInputs