Should the `language` for transformers.pipeline be set as `yue` instead of `zh`?
#8
by
JinchaoLove
- opened
As copied from part of the README
lang = 'zh'
pipe.model.config.forced_decoder_ids = pipe.tokenizer.get_decoder_prompt_ids(language=lang, task="transcribe")
# [(1, 50260), (2, 50359), (3, 50363)]
pipe.tokenizer.get_decoder_prompt_ids(language='yue', task="transcribe")
# [(1, 50358), (2, 50359), (3, 50363)]
Should the language be set as zh or yue, or either of them is okay?
yue is added for whisper v3 if I remember correctly, so I don't think yue would work