[Refactor] Move MM data parsing outside processor (#33408)

Signed-off-by: DarkLight1337 <tlleungac@connect.ust.hk>
This commit is contained in:
Cyrus Leung
2026-01-31 16:46:14 +00:00
committed by GitHub
parent 92924b2ddd
commit 88c3e114d8
43 changed files with 228 additions and 139 deletions
+7 -4
View File
@@ -262,11 +262,14 @@ class VoxtralDummyInputsBuilder(BaseDummyInputsBuilder[VoxtralProcessingInfo]):
)
res = tokenizer.mistral.encode_chat_completion(request)
dummy_tokens = res.tokens
# whixtral tokenizer adds padding to the audio
# so we need to update the audio arrays
dummy_mm_data["audio"] = [a.audio_array for a in res.audios]
return ProcessorInputs(prompt=dummy_tokens, mm_data=dummy_mm_data)
dummy_mm_inputs = self.info.parse_mm_data(
# whixtral tokenizer adds padding to the audio
# so we need to update the audio arrays
{**dummy_mm_data, "audio": [a.audio_array for a in res.audios]},
)
return ProcessorInputs(prompt=dummy_tokens, mm_items=dummy_mm_inputs)
class VoxtralMultiModalProcessor(BaseMultiModalProcessor[VoxtralProcessingInfo]):