(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
Korean-OpenOrca-13B is an auto-regressive language model based on the LLaMA2 transformer architecture.
Repo Link
Github Korean-OpenOrca: 🐳Korean-OpenOrca🐳
Base Model hyunseoki/ko-en-llama2-13b
Training Dataset
I use kyujinpy/KOR-OpenOrca-Platypus-v3.
(with NEFTune.)
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🐳] | 46.59 | 42.06 | 53.95 | 42.28 | 43.55 | 51.12 |
| KOR-Orca-Platypus-13B🐳-v2 | 49.48 | 44.03 | 54.43 | 42.23 | 41.64 | 65.05 |
| KOR-Orca-Platypus-13B🐳-v3 | 48.37 | 43.77 | 54.27 | 42.66 | 38.58 | 62.57 |
Compare with Top 4 SOTA models. (update: 10/09)
Implementation Code
### KO-Platypus
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch
repo = "kyujinpy/KOR-Orca-Platypus-13B-v3"
OpenOrca = AutoModelForCausalLM.from_pretrained(
repo,
return_dict=True,
torch_dtype=torch.float16,
device_map='auto'
)
OpenOrca_tokenizer = AutoTokenizer.from_pretrained(repo)