(주)미디어그룹사람과숲과 (주)마커의 LLM 연구 컨소시엄에서 개발된 모델입니다
The license is cc-by-nc-sa-4.0.
🐳KoR-Orca-Platypus-13B🐳
Model Details
Model Developers Kyujin Han (kyujinpy)
Input Models input text only.
Output Models generate text only.
Model Architecture
KoR-Orca-Platypus-13B is an auto-regressive language model based on the LLaMA2 transformer architecture.
Repo Link
Github Korean-OpenOrca: 🐳KoR-Orca-Platypus-13B🐳
Base Model hyunseoki/ko-en-llama2-13b
Training Dataset
Version of combined dataset: kyujinpy/KOR-OpenOrca-Platypus
I combined OpenOrca-KO and kyujinpy/KOpen-platypus. I use A100 GPU 40GB and COLAB, when trianing.
Model Benchmark
KO-LLM leaderboard
- Follow up as Open KO-LLM LeaderBoard.
| Model | Average | Ko-ARC | Ko-HellaSwag | Ko-MMLU | Ko-TruthfulQA | Ko-CommonGen V2 |
|---|---|---|---|---|---|---|
| KoR-Orca-Platypus-13B🐳(ours) | 50.13 | 42.06 | 53.95 | 42.28 | 43.55 | 68.78 |
| GenAI-llama2-ko-en-platypus | 49.81 | 45.22 | 55.25 | 41.84 | 44.78 | 61.97 |
| KoT-Platypus2-13B | 49.55 | 43.69 | 53.05 | 42.29 | 43.34 | 65.38 |
| KO-Platypus2-13B | 47.90 | 44.20 | 54.31 | 42.47 | 44.41 | 54.11 |
| Korean-OpenOrca-13B🐳 | 47.85 | 43.09 | 54.13 | 40.24 | 45.22 | 56.57 |
Compare with Top 4 SOTA models. (update: 10/14)
Implementation Code
### KO-Platypus
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
# model loading
repo = "kyujinpy/KoR-Orca-Platypus-13B"
OpenOrca = AutoModelForCausalLM.from_pretrained(
repo,
return_dict=True,
torch_dtype=torch.float16,
device_map='auto'
)
OpenOrca_tokenizer = AutoTokenizer.from_pretrained(repo)