Skip to content

Conversation

@qnguyen3
Copy link

What does this PR do?

Add ColQwen2_5 in 🤗 transformers.

Who can review?

Additional details

The newly converted model weights are stored in qnguyen3/colqwen2_5-v0.2-hf.

Progress checklist

  • Created script that successfully runs the forward() pass using the original repository and checkpoint
  • Successfully added the model skeleton to 🤗 Transformers
  • Successfully converted original checkpoint to 🤗 Transformers checkpoint
  • Successfully ran forward() pass in 🤗 Transformers that gives identical output to original checkpoint
  • Finished model tests in 🤗 Transformers
  • Successfully added tokenizer in 🤗 Transformers
  • Run end-to-end integration tests
  • Finished docs
  • Uploaded model weights to the Hub
  • Submitted the pull request

The implementation is production-ready and maintains full compatibility with the original ColQwen2.5 functionality. Looking forward to getting this merged! 🚀

Copy link
Member

@yonigozlan yonigozlan left a comment

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

Hey @qnguyen3, thanks for working on this! We just merged ColQwen2 in Transformers, I think it should be possible to use the colqwen2 modeling code with a colqwen2.5 checkpoint with minimal to no changes, so if we can avoid to add another model that would be great!
Can you try to make this work with the ColQwen2 modeling code as is? And if you find that some changes are needed, you can open a PR with those changes then :).
Thanks!

@yonigozlan
Copy link
Member

Closing this for the time being following my previous comment

@sahil-kabir
Copy link
Contributor

Hi @yonigozlan, I'd love to take a shot at this. I'm going to upload ColQwen2.5 to the hub under my account for now, and then test if it can load directly with the existing ColQwen2 modeling code by adjusting the config. At the very worst, it might need some minor tweaks in the code for ColQwen2. I'll open a WIP PR if that sounds good to you.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

5 participants