SiWeiTu

I also try to use llama3-8b as llm, and i add the code "tokenizer.pad_token_id=-100", after that when i start pretrain, i get the error: OverflowError: out of range integral type...

It seems that the model , 'damo/speech_paraformer-large_asr_nat-zh-cn-16k-common-vocab8404-online,' is able to successfully recognize test recordings provided by you and other synthesized audio, but when i try to recognize my own recorded...

Hello，do you solve the problem？

https://github.com/MCG-NJU/VideoMAE/issues/9

the same error

I encounter the same question when ds version==0.14，any solution to solve it？Must I change the version？

> How to obtain as a result the masks without the background image? do you have any way to relize it？

> The author updated on this page：#201 i have try  thanks， but i need to retain the main object in the image instead of black color...

the result is too bad to accept...