Ao Tang

Results 25 comments of Ao Tang

> Please capture the error and include in the ImportError so users will know exactly what's not being imported properly. ImportError is raised at the model level not layer level....

Hi @ArthurZucker , I updated the PR with the changes @suhara and @gheinrich had before. Let me know how you want to proceed. Thanks!

@ArthurZucker I have some question relating to the `Copied from` convention, if the code is not exactly the same but adapted from it, should I still have the `Copied from`...

@ArthurZucker All the issues has been resolved. Please take a look again! Thanks a lot for the review! Looks like there is still some CI not passed. I'll address them...

> IMO the best thing is to use `PreTokenizerFast` But `PreTrainedTokenizerFast` does not support directly load setencepiece (`tokenizer.model`) file.

> > But PreTrainedTokenizerFast does not support directly load setencepiece (tokenizer.model) file. > > It does not, but we can simply convert it in the conversion script! It's really up...

Hi @riyajatar37003 , if you convert the NeMO ckpt to hf ckpt using the `export_ckpt`, you should be able to run inference on HF side using script similar to below:...

Thanks @xiangxu-google for the PR. I will try to verify this on our end as well

Hi @xiangxu-google , I wonder have you tested the inference and compare the results between NeMo's and HF's ? I observed that for 1B model at least, NeMo output is...

Thanks @xiangxu-google , could you summarize what you have changed ? It is hard to see your change with the force merge. I also managed to review your change and...