[Bugfix] Fix Whisper tokenization (#34011)

Signed-off-by: NickLucche <nlucches@redhat.com>

vllm/model_executor/models/whisper.py

@@ -727,6 +727,14 @@ class WhisperMultiModalProcessor(EncDecMultiModalProcessor[WhisperProcessingInfo
                 **mm_kwargs,
                 sampling_rate=feature_extractor.sampling_rate,
             )
+        # The HF WhisperProcessor passes **kwargs to both the tokenizer
+        # and the feature extractor. Text-tokenizer kwargs like
+        # `truncation` and `max_length` must be removed when audio data
+        # is present, otherwise the feature extractor interprets
+        # `max_length` as raw audio samples and truncates the audio.
+        tok_kwargs = {
+            k: v for k, v in tok_kwargs.items() if k not in ("truncation", "max_length")
+        }
         processed_outputs = super()._call_hf_processor(
             prompt=prompt,
             mm_data=mm_data,