Was this also trained on 7000 steps?
#1
by
Xianjun
- opened
Hi,
I noticed in another link https://huggingface.co/EleutherAI/sae-llama-3-8b-32x-v2 you mentioned that they were trained on 7000 steps. Were this repo also trained on 7000 steps?
It was trained for over 30k steps on over 8B tokens.