llama-1b-hf-32768-fpf

作者 mesolitica · 下载 204 · 点赞 0 · 更新 2023-11-04

Full Parameter Finetuning 1B 32768 context length Llama2 on Malaysian text

1B derived from first 4 layers 7B model.

README at https://github.com/mesolitica/malaya/tree/5.1/session/llama2#1b-32768-context-length-flash-attention-2

WandB, https://wandb.ai/mesolitica/fpf-Llama-2-1b-32k-hf