Vivek Iyer

Hello, thank you so much for responding. I launch it as a standalone job like this: ``` CUDA_VISIBLE_DEVICES=0,1,2,3 python ${preprocessing_dir}/query_bloom.py \ --name bigscience/bloom --dtype int8 \ --batch_size 1 --num-beams 1...

Hi @richarddwang , yes but I do set max_new_tokens to be 64 (L20 in the gist). So this does not seem to be the issue

I have tried all the approaches in this thread (including @gegallego, @joelb-git, @alexandra-chron) and none of them work for me. I still get dimension mismatch errors of this sort: `copying...

Thank you so much for providing the snippets, @joelb-git! I tried to replicate them exactly, by cloning the mentioned commit and reproducing your changes. The issue I mentioned got resolved,...

@joelb-git thank you so much for going to the effort of downloading the huge mBART binary file, and running cmp on it to verify its the same. 1. Yes, I...

Hi @gegallego and @joelb-git Thanks so much for both of your help! Providing `ML50_langs.txt` as `lang-dict` did work after all, and my model is training successfully :) Perhaps the README...

Hi, is there any update on this? I'm interested too! @pluesclues @danielhanchen