pythia-350m-deduped is bugged

#1
by MaxLohMusic - opened

The output is gibberish as if the tokenizer did the tokens all wrong. I found the same exact code works perfectly fine with pythia-125m-deduped

Edit: Additionally confirmed that "pythia-350m" is working perfectly fine, so only the 350m deduped version is bugged

EleutherAI org

This looks like the same problem as is reported here: https://huggingface.co/EleutherAI/pythia-1.3b-deduped/discussions/1#638f8655dd47b2ac3d715f65

We are looking into it.

EleutherAI org

This has been resolved.

stellaathena changed discussion status to closed

Sign up or log in to comment