[Perf] Improve multimodal item handling from O(n) to O(log n) per step (#44212)

Signed-off-by: Andy Lo <andy@mistral.ai>
This commit is contained in:
Andy Lo
2026-06-03 11:00:26 +00:00
committed by GitHub
parent 1fa9ea09f6
commit 95b1615ec9
6 changed files with 92 additions and 44 deletions
@@ -44,7 +44,7 @@ class MistralCommonFeatureExtractor:
if not self.audio_encoder.audio_config.is_streaming:
audio = self.audio_encoder.pad(audio, self.sampling_rate)
audios_processed.append(torch.tensor(audio))
audios_processed.append(torch.from_numpy(audio))
return BatchFeature(
{"audio_arrays": audios_processed}, tensor_type=return_tensors