Spaces:
Running
Running
LLaMA-2-13b chat
#44
by
levmckinney
- opened
Adds a tuned lens trained on llama-2-13b-chat-hf with togethercomputer/RedPajama-Data-1T-Sample and default hyper parameters for tuned-lens==0.2.0 wandb. While this will be the default, in addition, we provide a lens trained on Anthropic/hh-rlhf note that this lens is not trained for the full 250 steps since Anthropic/hh-rlhf is too small wandb
levmckinney
changed pull request status to
open
levmckinney
changed pull request title from
LLaMA 13b chat
to LLaMA-2-13b chat
levmckinney
changed pull request status to
merged