science-of-finetuning/replaced-token-chat-template-Meta-Llama-3.1-8B-L16-mu2.0e-02-lr1e-04-local-shuffling-CCLoss

wandb run: https://wandb.ai/jkminder/crosscoder/runs/kj2108sr?nw=nwuserjkminder

Trained on replaced activations for chat model (not the default template, butthe one modififed with the replace token map)