werty1248/Llama-3-Ko-8B-OpenOrca

Llama-3-Ko-OpenOrca

Model Details

Model Description

Original model: beomi/Llama-3-Open-Ko-8B

Dataset: kyujinpy/OpenOrca-KO

Training details

Training: Axolotl을 이용해 LoRA-8bit로 4epoch 학습 시켰습니다.

• sequence_len: 4096
• bf16

학습 시간: A6000x2, 6시간

Evaluation

• 0 shot kobest

Tasks	n-shot	Metric	Value		Stderr
kobest_boolq	0	acc	0.5021	±	0.0133
kobest_copa	0	acc	0.6920	±	0.0146
kobest_hellaswag	0	acc	0.4520	±	0.0223
kobest_sentineg	0	acc	0.7330	±	0.0222
kobest_wic	0	acc	0.4881	±	0.0141

• 5 shot kobest

Tasks	n-shot	Metric	Value		Stderr
kobest_boolq	5	acc	0.7123	±	0.0121
kobest_copa	5	acc	0.7620	±	0.0135
kobest_hellaswag	5	acc	0.4780	±	0.0224
kobest_sentineg	5	acc	0.9446	±	0.0115
kobest_wic	5	acc	0.6103	±	0.0137

License:

https://llama.meta.com/llama3/license