whisper-large-v3 (in demo code) VS whisper-large-v2 (in kaggle notebook)

[Binah-CS]

@sanchit-gandhi - the demo is using the v3 dataset, but the kaggle notebook and readme - all reference v2.
In our tests, the v3 give significantly better output for our test audio files than v2, but if I try to update the notebook to v3, it doesn't even compile (v2 compiled just fine). Can you point me in the right direction?

---

ValueError Traceback (most recent call last)
Cell In[7], line 4
1 from whisper_jax import FlaxWhisperPipline
2 import jax.numpy as jnp
----> 4 pipeline = FlaxWhisperPipline("openai/whisper-large-v3", dtype=jnp.bfloat16, batch_size=16)

File /usr/local/lib/python3.8/site-packages/whisper_jax/pipeline.py:82, in FlaxWhisperPipline.init(self, checkpoint, dtype, batch_size, max_length)
79 self.checkpoint = checkpoint
80 self.dtype = dtype
---> 82 self.processor = WhisperProcessor.from_pretrained(self.checkpoint)
83 self.feature_extractor = self.processor.feature_extractor
84 # potentially load fast tokenizer if available

File /usr/local/lib/python3.8/site-packages/transformers/processing_utils.py:226, in ProcessorMixin.from_pretrained(cls, pretrained_model_name_or_path, cache_dir, force_download, local_files_only, token, revision, **kwargs)
223 if token is not None:
224 kwargs["token"] = token
--> 226 args = cls._get_arguments_from_pretrained(pretrained_model_name_or_path, **kwargs)
227 return cls(*args)

File /usr/local/lib/python3.8/site-packages/transformers/processing_utils.py:270, in ProcessorMixin._get_arguments_from_pretrained(cls, pretrained_model_name_or_path, **kwargs)
267 else:
268 attribute_class = getattr(transformers_module, class_name)
--> 270 args.append(attribute_class.from_pretrained(pretrained_model_name_or_path, **kwargs))
271 return args

File /usr/local/lib/python3.8/site-packages/transformers/tokenization_utils_base.py:1854, in PreTrainedTokenizerBase.from_pretrained(cls, pretrained_model_name_or_path, cache_dir, force_download, local_files_only, token, revision, *init_inputs, **kwargs)
1851 else:
1852 logger.info(f"loading file {file_path} from cache at {resolved_vocab_files[file_id]}")
-> 1854 return cls._from_pretrained(
1855 resolved_vocab_files,
1856 pretrained_model_name_or_path,
1857 init_configuration,
1858 *init_inputs,
1859 token=token,
1860 cache_dir=cache_dir,
1861 local_files_only=local_files_only,
1862 _commit_hash=commit_hash,
1863 _is_local=is_local,
1864 **kwargs,
1865 )

File /usr/local/lib/python3.8/site-packages/transformers/tokenization_utils_base.py:2066, in PreTrainedTokenizerBase._from_pretrained(cls, resolved_vocab_files, pretrained_model_name_or_path, init_configuration, token, cache_dir, local_files_only, _commit_hash, _is_local, *init_inputs, **kwargs)
2062 current_index = len(tokenizer) + len(tokens)
2063 if has_tokenizer_file and index != current_index and tokenizer.convert_tokens_to_ids(token) != index:
2064 # Tokenizer fast: added token needs to either be in the vocabulary with the proper index or the
2065 # index is the current length of the tokenizer (not in vocabulary)
-> 2066 raise ValueError(
2067 f"Wrong index found for {token}: should be {tokenizer.convert_tokens_to_ids(token)} but found "
2068 f"{index}."
2069 )
2070 elif not has_tokenizer_file and index != current_index:
2071 # Tokenizer slow: added token cannot already be in the vocabulary so its index needs to be the
2072 # current length of the tokenizer.
2073 raise ValueError(
2074 f"Non-consecutive added token '{token}' found. "
2075 f"Should have index {current_index} but has index {index} in saved vocabulary."
2076 )

ValueError: Wrong index found for <|0.02|>: should be None but found 50366.

Thanks