Hi All - This is my first time using such a large model on my local computer and I’m running into issues actually downloading it. I’ve tried a variety of approaches:
- I’ve tried downloading/installing the model with model = AutoModelForSeq2SeqLM.from_pretrained(“facebook/nllb-moe-54b”, force_download=True, resume_download=True). The last time I tried this, I would keep getting timeout errors which I was able to get around with resume_download set to true. This seemed to be working until it threw an error for file sizes not matching. Then when it restarted, it started an entirely new download (after I had already downloaded about 30GB of the model). I saved this by the way so if there’s any way to get it to start over correctly with this, that would be preferred.
- I’ve tried using huggingface-cli but it does not appear to have a “download” option on Windows so I’m kind of stumped there.
- I’m currently the in process of trying to clone the repo with Windows git but so far it seems like it may have timed out.
I could try just downloading the files from the repo directly but then how would I get them in the correct file structure and ensure I can actually call the model from my notebook? Any ideas on how I can brute-force this? I’ve been playing with this all weekend and it’s starting to get kind of annoying. Thanks. Any help is greatly appreciated!