AlignmentResearch / tuned-lens

Tools for understanding how transformer predictions are built layer-by-layer
https://tuned-lens.readthedocs.io/en/latest/
MIT License
438 stars 47 forks source link

Upgrade to support torch 2.0 #14

Closed levmckinney closed 1 year ago

levmckinney commented 1 year ago

This would make our dependences a lot simpler plus we might get a nice performance improvement.

levmckinney commented 1 year ago

The priority of this task has increased substantially due to https://github.com/pytorch/pytorch/issues/89817. This bug is currently not patched in 1.13. It's not clear if it ever will be, and it is blocking me from loading pythia and GPTNeo models on Hofvarpnir.