Modelpublic

mesolitica/llama-2b-hf-32768-fpf

sourceHugging Faceupdated 3y agoView on Hugging Face

1likes257downloads

Model Card

Full Parameter Finetuning 2B 32768 context length Llama2 on Malaysian text

2B derived from first 5 layers 13B model.

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

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