ko-en-llama2-13b-mixed-v2

The license is cc-by-nc-2.0.

GAI-LLM/ko-en-llama2-13b-mixed-v2

Model Details

Model Developers Donghoon Oh, Hanmin Myung, Eunyoung Kim (SK C&C G.AI Eng)

Input Models input text only.

Output Models generate text only.

Model Architecture
GAI-LLM/ko-en-llama2-13b-mixed-v2 is an auto-regressive language model based on the LLaMA2 transformer architecture.

Base Model hyunseoki/ko-en-llama2-13b

Training Dataset

• We combined Open Korean Dateset using mixed-strategy.
  • Kopen-platypus + Everythinglm v2 + koalpaca_v1.1 + koCoT2000
• We use A100 GPU 80GB * 8, when training.

Model Benchmark

KO-LLM leaderboard

• Follow up as Open KO-LLM LeaderBoard.

Implementation Code

### GAI-LLM/ko-en-llama2-13b-mixed-v2
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

repo = "GAI-LLM/ko-en-llama2-13b-mixed-v2"
model = AutoModelForCausalLM.from_pretrained(
        repo,
        return_dict=True,
        torch_dtype=torch.float16,
        device_map='auto'
)
tokenizer = AutoTokenizer.from_pretrained(repo)