epfLLM/Megatron-LLM

Add falcon support in megatron2hf.py

Closed this issue · 4 comments

AleHD commented
Add falcon support in megatron2hf.py

Is someone already working on this? Otherwise you could assign me please.

AleHD commented

Nobody is working on it at the moment, go for it

I implemented the basic megatron to HF convertion for falcon7b-like models. See https://gist.github.com/malteos/85fd117cb0ba4cd28882464026252ee9

The implementation is still pretty much WIP and requires manually downloading the RWModel + RWConfig files from HF hub.

very cool, thanks! would you like to make a PR out of it and then one of @andreaskoepf @Olivia-fsm @mpagli @AleHD could have a look?