Model fails with error about missing generation config
When this model is used in Buzz application https://github.com/chidiwilliams/buzz/ it fails with the following error.
ValueError: You are trying to return timestamps, but the generation config is not properly set. Make sure to initialize the generation config with the correct attributes that are needed such as
no_timestamps_token_id. For more details on how to generate the approtiate config, refer to https://github.com/huggingface/transformers/issues/21878#issuecomment-1451902363
Please consider adding generation config to the model repository
I added the generation config as I would assume it to be the same as in this repo: https://huggingface.co/Finnish-NLP/whisper-large-finnish-v3-ct2
Let me know if it is ok
Thanks @RASMUS works like a charm now.
Maybe I can also kindly ask you to add the generation config also to https://huggingface.co/Finnish-NLP/whisper-large-v2-finnish
Hmm now when I look into it I think there might be some issues in the generation config:
Fi is:
"<|fi|>": 50277,
but in the config I have now:
"forced_decoder_ids": [
[
1,
null
],
[
2,
50360
]
],
Shouldn't that 2nd value be 50277?
I need to take a look into this later maybe. I have been planning to do the finetuning on more "balanced" dataset
Maybe it still works.
I did not see errors in the training code.
Tested locally and seems to still work
Seems to work for me as well