[Bug Report] Updates to Gemma

[cmathw]

Since support for Gemma was merged March 14th, there have been a number of changes to the upstream HF model file such that we no longer have good agreement across logits and cache. A large chunk of these changes exist here but I think there are also more recent relevant changes in other commits. I will be opening a corresponding PR to address these issues tomorrow, in the meantime I would not trust Gemma outputs and activations.

Checklist

• I have checked that there is no similar issue in the repo (required)