Clément Dumas
Clément Dumas
closing this as there is now a https://nnsight.net/notebooks/tutorials/diffusion_lens/ tutorial
Are you using the latest version on github ? There was some issues with llama models that were fixed. You can update with ` pip install --force-reinstall --no-deps git+https://github.com/neelnanda-io/TransformerLens/`
I think you can do two different traces instead of 2 invokes in your current code and it should work as expected
Hey, If you look at the token probability distribution, they should be the same. TransformerLens performs many operations under the hood to facilitate interpretability (see [this explanation](https://github.com/TransformerLensOrg/TransformerLens/blob/main/further_comments.md#weight-processing)). If you don't...
Hey, this feature would be super helpful for automated workflows!
@JadenFiotto-Kaufman I can try to add tests if needed but I wanted to first check if you were ok with adding this feature to the nnsight core.
Hmm, would you still pass the alias dict through the NNSight class init? I guess one advantage of renaming over aliasing is that when you print the model, you actually...
@JadenFiotto-Kaufman bump
Implemented in #298
@JadenFiotto-Kaufman thanks first bug resolved! However this other code also fails. This means that accessing a children of a renamed module to apply it during a trace doesn't work ```py...