Hugging Face's logo

tiiuae
/
falcon-40b

Text Generation
Transformers
PyTorch
Safetensors
falcon
custom_code
text-generation-inference
Model card Files
xet
 Community
117

Revert in-library PR

#90
by Rocketknight1 HF Staff - opened
base: refs/heads/main
←
from: refs/pr/90
Discussion Files changed
+359
-590
Rocketknight1
No description provided.
Revert "Move to in-library checkpoint (#81)"46f48b52
Rocketknight1 changed pull request status to open
sgugger changed pull request status to merged
borzunov

Hi @Rocketknight1 @sgugger ,

Thanks for your work on Falcon support in transformers!

Could you please share if this model is going to be converted to the new in-library format (model_type="falcon") soon? If yes, when do you expect this to happen?

I'm asking because I'm working on Falcon support in Petals (https://github.com/bigscience-workshop/petals) and would like to understand (1) which format we need to support
and (2) if we can make a release with Falcon this week or should wait until the format changes.

I'd really appreciate you sharing your plans on this matter!
sgugger

The model will be converted back to the new in-library format soon. The problem is that the library test-generation-inference added support for Falcon by checking its model_type and we need to wait for a new release of TGI that supports both model types (falcon and the current RefinedWeb) to avoid breaking many users deployment of Falcon through TGI.
borzunov

Got it, thanks!

Sign up or log in to comment