File size: 1,100 Bytes
5a82505
d48b662
 
 
5a82505
d48b662
b9e3ca2
d48b662
 
d8b43c8
57506eb
 
d48b662
d8b43c8
 
b9e3ca2
 
 
 
 
 
 
 
 
 
d8b43c8
d48b662
30229c0
 
57506eb
 
d48b662
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
---
license: cc-by-nc-4.0
base_model:
- google/gemma-2-2b-it
---

# Gemma-2-2b ์ดˆ๋“ฑํ•™์ƒ ๊ธ€ ๋ณ€ํ™˜๊ธฐ


## Model Description
- Gemma2-2b ํ•œ๊ตญ ์ดˆ๋“ฑํ•™์ƒ ๊ธ€ ๋ณ€ํ™˜๊ธฐ๋Š” ์ž…๋ ฅํ•œ ๊ธ€์„ ์ดˆ๋“ฑํ•™์ƒ์ด ์“ด ๊ฒƒ ๊ฐ™์€ ๊ธ€๋กœ ๋ณ€ํ™˜ํ•ด ์ฃผ๋Š” ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค.
- Gemma2-2b-it ๋ชจ๋ธ์„ base model๋กœ ์‚ฌ์šฉํ•˜์˜€๊ณ  LoRA๊ธฐ๋ฒ•์„ ์‚ฌ์šฉํ•˜์—ฌ ํšจ์œจ์ ์œผ๋กœ fine-tuning ํ•˜์˜€์Šต๋‹ˆ๋‹ค. 

## Usage
```
import transformers
import torch
from huggingface_hub import notebook_login

notebook_login()

BASE_MODEL = "skwh54/Gemma-2-2b-it-elementary-style-document"

model = AutoModelForCausalLM.from_pretrained(BASE_MODEL, device_map={"":0})
tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)
```

## Training data
- [korean_smile_style_dataset](https://github.com/smilegate-ai/korean_smile_style_dataset)์„ ์‚ฌ์šฉํ•˜์˜€์Šต๋‹ˆ๋‹ค.
- ๋ณธ ๋ฐ์ดํ„ฐ๋Š” Smilegate AI์—์„œ ๊ณต๊ฐœํ•˜๋Š” ํ•œ๊ตญ์–ด ๋ฌธ์ฒด ์Šคํƒ€์ผ ๋ณ€ํ™˜ "SmileStyle" ๋ฐ์ดํ„ฐ์…‹์ž…๋‹ˆ๋‹ค.
- ์—ฌ๋Ÿฌ ์Šคํƒ€์ผ์˜ ๋ฌธ์ฒด๊ฐ€ ์กด์žฌํ•˜๋ฉฐ ์ด ์ค‘์—์„œ formal๊ณผ choding ์Šคํƒ€์ผ์˜ ๋ฐ์ดํ„ฐ๋งŒ์„ ์ถ”์ถœํ•˜์—ฌ ์‚ฌ์šฉํ•˜์˜€์Šต๋‹ˆ๋‹ค.