Alex McKinney
Alex McKinney
Yeah, it is correct behaviour - what I meant though is that I did have a `LLAMA_INPUTS_DOCSTRING` in a previous commit, but running `make fix-copies` overwrote this docstring with the...
@sanchit-gandhi fixed the CI issue (I think) by just adding more `Copied from ...` comments and deleting the class level comment. I also fixed the merge conflict. We should be...
@sanchit-gandhi the CI still fails, this is for two reasons. Could you assist me with resolving this? 1. The documentation test fails as it tries to load the checkpoint `_CHECKPOINT_FOR_DOC`...
> Could you open a pull request on the HF Hub to add the Flax model weights to the checkpoint? PR is open 🙂 Git LFS is a great thing....
Hey @sanchit-gandhi, thanks for the second review~ I can try with Llama 2 weights, however last time I requested access I never got a response. I will try again when...
Hey @sanchit-gandhi, thanks for bearing with me. I have addressed your comments. They were quite small but I didn't have the headspace to think about this earlier 😅 Regarding Llama...
@sanchit-gandhi I was also thinking of adding a Flax version of LLama (and also GPT-NeoX, maybe others) as some Flax practice. I couldn't find a guide on adding a new...
Thanks @sanchit-gandhi that was very comprehensive! I'll let you know how I get on. :hugs:
Got a bit caught up with real life stuff, but I will be working on this more intensively from Monday, aiming to finish something by end of week.
@sanchit-gandhi I made a draft PR of my current progress, see #24587. Sorry, I haven't made the full model, been very busy 😓