Marcin Kardas

Changing pytorch sharing strategy seems to help: ```python import torch torch.multiprocessing.set_sharing_strategy("file_system") ```

Hi @fengyunzaidushi, thanks for reporting. What is your `bleach` version?

Thanks @fengyunzaidushi for the details. I updated the bleach version in requirements.txt (~=5.0.1), can you reinstall bleach and check again?

@mso1830 I pushed a fix to the main branch, not yet on pypi. Can you try it if it works for you?

Can you check if your `prompt` is shorter than 2047 tokens?

You can run `python -m torch.utils.collect_env` as well as `pip list`. What's the prompt's length in tokens now? By "ran for fewer more iterations" do you mean with the exact...

Hi @BruceStayHungry, we fixed the 30B checkpoint. Can you check if it works now?

What are your `galai` and `transformers` versions? What do you get for: ```python from transformers import AutoTokenizer tokenizer = AutoTokenizer.from_pretrained("facebook/galactica-120b") ``` ?

It looks like the model cannot fit into the RAM + VRAM. What is the amount of free RAM & VRAM? You can reduce memory requirement by using float16 inference...

It seems that either your `pytorch` installation is broken or that you're running `python` from the same directory where you downloaded `pytorch`'s source code. 1. can you check if your...