Adding `safetensors` variant of this model

#5
by SFconvertbot - opened

This is an automated PR created with https://huggingface.co/spaces/safetensors/convert

This new file is equivalent to pytorch_model.bin but safe in the sense that
no arbitrary code can be put into it.

These files also happen to load much faster than their pytorch counterpart:
https://colab.research.google.com/github/huggingface/notebooks/blob/main/safetensors_doc/en/speed.ipynb

The widgets on your model page will run using this model even if this is not merged
making sure the file actually works.

If you find any issues: please report here: https://huggingface.co/spaces/safetensors/convert/discussions

Feel free to ignore this PR.

Friendly ping to @lysandre and @joaogante . Is it safe to merge this PR?

@osanseviero This is our own bot, so it should be! I see no limitations of gpt2 + safetensors from the transformers side

OpenAI community org

Verified that the two checkpoints had equal layers of equal values, merging!

lysandre changed pull request status to merged
OpenAI community org

YAYYYY

gpt2-xl is now safetensors activated

image.png

@lysandre we can automate that check, our PT -> TF script does it! I'm going to open a PR today (mostly copy/paste), so it can free the team some more time ⏳

OpenAI community org

The safetensors Space also does it, but for super widely used checkpoints like this I find it important to double check 😀

In particular an inference test is super important, as safetensors tied weights are not managed the same way as pyorch's bin weights.

See #26292 and #26422 which were necessary after merging safetensors weights.

If your script would have prevented these from happening, would love to automate it!

Thanks

As discussed on slack: the bot does double-check the inference code with the right architecture (here), but it's not checking the hidden-states. The hidden-states part will be added :)

Sign up or log in to comment