Bryce Meyer

Results 138 comments of Bryce Meyer

[Proposal] Expand quantization model support

It was not done due to the person who added it being a volunteer. We can definitely put expanding this on the list of todos.

[Proposal] Add support for TracrBench

Can you add a link to the models on HuggingFace, and a link to the source code? Most likely, you will be able to utilize the majority of existing component...

[Bug Report] Gemma-2-2b-it output logit doesn't match with huggingface

@yanivnik What version of TransformerLens were you using? The latest version should have modified that, but that does not mean it is now 100% accurate. It would be helpful to...

Force unsupported models

This doesn't entirely align with the direction of transformer lens the way this PR is currently structured. The roadmap for TransformerLens 4.0 is to add a number of hooks for...

[Proposal] Compatibility for OLMo and OLMo2?

It should be in the next release!

removed einsum causing error when use_atten_result is enabled

Sorry @oliveradk! The last month has been pretty busy, and this one slipped through the cracks. Looks perfect, and it will be in the next release.

NanoGPT Conversation did not handle case when there were no biases in model

Thanks for finding this, and fixing it for everyone. There is just one type error that needs to be resolved when creating your tensor for `mlp.b_in` in your else block....

[Proposal] Remove the overhead caused by full_hook.__name__ = (hook.__repr__())?

If you have time to submit a PR for this, I would be happy to accept a change that gives an option to `add_hook` to skipping the name. Maybe something...