finetune.run failed on assert batch.text_to_units.prev_output_tokens is not None

facebookresearch / seamless_communication

Foundational Models for State-of-the-Art Speech and Text Translation

Other

10.77k stars 1.05k forks source link

finetune.run failed on assert batch.text_to_units.prev_output_tokens is not None #351

Open laleye opened 7 months ago

laleye commented 7 months ago

I tried to finetune on a new language using m4t_cli scripts without success. I have the following error which I cannot understand. However, it is indicated in the dataloader thattext_to_units.prev_output_tokens must be None

--> 119 assert batch.text_to_units.prev_output_tokens is not None
    120 dummy_context = contextmanager(lambda: iter([None]))()
    121 with torch.no_grad() if self.freeze_t2u else dummy_context:  # type:ignore

AssertionError:

Help please!

mhlakhani commented 4 months ago

@laleye : https://github.com/facebookresearch/seamless_communication/issues/424 suggests this might come when source and target languages are the same. I'm also hitting this exact error in that case. Are you also running into this issue in the same context?

Huzaifah-AI commented 2 months ago

I am training for Text to speech do I have to provide a target language?

NMikaa commented 2 months ago

hello, did you fix this? how?