README.md

---
language:
- ko
datasets:
- kyujinpy/OpenOrca-ko-v3
library_name: transformers
pipeline_tag: text-generation
license: cc-by-nc-sa-4.0
---
**(주)미디어그룹사람과숲과 (주)마커의 LLM 연구 컨소시엄에서 개발된 모델입니다**  
**The license is `cc-by-nc-sa-4.0`.**  

# **🐳Korean-OpenOrca-13B-v2🐳**  
![img](./Korean-OpenOrca.png)  

## Model Details

**Model Developers** Kyujin Han (kyujinpy)

**Model Architecture**  
Korean-OpenOrca-13B is an auto-regressive language model based on the LLaMA2 transformer architecture.

**Repo Link**  
Github Korean-OpenOrca: [🐳Korean-OpenOrca🐳](https://github.com/Marker-Inc-Korea/Korean-OpenOrca)  

**Base Model**  [hyunseoki/ko-en-llama2-13b](https://huggingface.co/hyunseoki/ko-en-llama2-13b)   

**Training Dataset**  
I use [OpenOrca-ko-v3](https://huggingface.co/datasets/kyujinpy/OpenOrca-ko-v3).  
Using DeepL, translate about [OpenOrca](https://huggingface.co/datasets/Open-Orca/OpenOrca).

I use A100 GPU 40GB and COLAB, when trianing.

# Model comparisons
| Model | Average |Ko-ARC | Ko-HellaSwag | Ko-MMLU | Ko-TruthfulQA | Ko-CommonGen V2 |
| --- | --- | --- | --- | --- | --- | --- |
| [Korean-OpenOrca-13B🐳] | 48.79 | 43.09 | 54.13 | 40.24 | 45.22 | 61.28 |  
| [Korean-OpenOrca-13B-v2🐳] | 48.17 | 43.17 | 54.51 | 42.90 | 41.82 | 58.44 | 
| Korean-OpenOrca-13B-v3🐳 | 48.86 | 43.77 | 54.30 | 41.79 | 43.85 | 60.57 | 

# Implementation Code
```python
### KO-Platypus
from transformers import AutoModelForCausalLM, AutoTokenizer
import torch

repo = "kyujinpy/Korean-OpenOrca-13B-v3"
OpenOrca = AutoModelForCausalLM.from_pretrained(
        repo,
        return_dict=True,
        torch_dtype=torch.float16,
        device_map='auto'
)
OpenOrca_tokenizer = AutoTokenizer.from_pretrained(repo)
```

---