[Bugfix] Fix multi-modal processors for transformers 4.48 (#12187)

This commit is contained in:
Cyrus Leung
2025-01-18 19:16:34 -08:00
committed by GitHub
parent 4e94951bb1
commit 630eb5b5ce
6 changed files with 198 additions and 35 deletions
+1 -8
View File
@@ -137,7 +137,7 @@ class UltravoxMultiModalProcessor(
mm_kwargs: Mapping[str, object],
) -> BatchFeature:
# Text-only input not supported in composite processor
if not mm_data:
if not mm_data or not mm_data.get("audios", []):
prompt_ids = self.info.get_tokenizer().encode(prompt)
prompt_ids = self._apply_hf_processor_tokens_only(prompt_ids)
return BatchFeature(dict(input_ids=[prompt_ids]), tensor_type="pt")
@@ -146,13 +146,6 @@ class UltravoxMultiModalProcessor(
audios = mm_data.pop("audios", [])
assert isinstance(audios, list)
if not audios:
return super()._call_hf_processor(
prompt=prompt,
mm_data=mm_data,
mm_kwargs=mm_kwargs,
)
feature_extractor = self.info.get_feature_extractor()
mm_kwargs = dict(
**mm_kwargs,