File size: 3,143 Bytes
9af5971
 
 
 
 
 
 
 
 
 
 
 
 
 
 
8020a08
 
 
 
 
ac6f1c0
8020a08
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
9af5971
8020a08
b9a0e2c
 
9af5971
 
 
 
 
 
 
b9a0e2c
 
9af5971
 
6c55003
 
 
 
 
 
 
 
 
 
 
 
d157774
 
 
9486288
 
 
 
 
 
 
 
 
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
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
---
license: llama3.2
datasets:
- CarrotAI/Carrot
- CarrotAI/Chat-Template
language:
- ko
- en
base_model:
- CarrotAI/Llama-3.2-Rabbit-Ko-3B-Instruct
pipeline_tag: text-generation
---

![image/webp](https://cdn-uploads.huggingface.co/production/uploads/64633ebb39359568c63b52ad/Cji1GoDqERF0zCR9gKJKV.webp)

## Model Description

### Model Details
- **Name**: Carrot Llama-3.2 Rabbit Ko 2412
- **Version**: 3B Instruct
- **Base Model**: CarrotAI/Llama-3.2-Rabbit-Ko-3B-Instruct
- **Languages**: Korean, English
- **Model Type**: Large Language Model (Instruction-tuned)

### Training Process
๋ณธ ๋ชจ๋ธ์€ ๋‹ค์Œ๊ณผ ๊ฐ™์€ ์ฃผ์š” ํ›ˆ๋ จ ๋‹จ๊ณ„๋ฅผ ๊ฑฐ์ณค์Šต๋‹ˆ๋‹ค:

1. **SFT (Supervised Fine-Tuning)**
   - ๊ณ ํ’ˆ์งˆ ํ•œ๊ตญ์–ด ๋ฐ ์˜์–ด ๋ฐ์ดํ„ฐ์…‹์„ ์‚ฌ์šฉํ•˜์—ฌ ๊ธฐ๋ณธ ๋ชจ๋ธ์„ ์„ธ๋ถ€ ์กฐ์ •

2. **DPO (Direct Preference Optimization)**
   - ์ธ๊ฐ„์˜ ์„ ํ˜ธ๋„๋ฅผ ์ง์ ‘์ ์œผ๋กœ ๋ฐ˜์˜ํ•˜์—ฌ ๋ชจ๋ธ์˜ ์‘๋‹ต ํ’ˆ์งˆ ๊ฐœ์„ 

### Limitations
- 3B ํŒŒ๋ผ๋ฏธํ„ฐ ๊ทœ๋ชจ๋กœ ์ธํ•œ ๋ณต์žกํ•œ ์ž‘์—…์—์„œ์˜ ์ œํ•œ์  ์„ฑ๋Šฅ
- ํŠน์ • ๋„๋ฉ”์ธ์— ๋Œ€ํ•œ ๊นŠ์ด ์žˆ๋Š” ์ „๋ฌธ์„ฑ ๋ถ€์กฑ
- ํŽธํ–ฅ์„ฑ ๋ฐ ํ™˜๊ฐ ๊ฐ€๋Šฅ์„ฑ

### Ethics Statement
๋ชจ๋ธ ๊ฐœ๋ฐœ ๊ณผ์ •์—์„œ ์œค๋ฆฌ์  ๊ณ ๋ ค์‚ฌํ•ญ์„ ์ตœ๋Œ€ํ•œ ๋ฐ˜์˜ํ•˜์˜€์œผ๋‚˜, ์‚ฌ์šฉ์ž๋Š” ํ•ญ์ƒ ๊ฒฐ๊ณผ๋ฅผ ๋น„ํŒ์ ์œผ๋กœ ๊ฒ€ํ† ํ•ด์•ผ ํ•ฉ๋‹ˆ๋‹ค.

### How to Use
```python
from transformers import AutoModelForCausalLM, AutoTokenizer

model = AutoModelForCausalLM.from_pretrained("CarrotAI/Llama-3.2-Rabbit-Ko-3B-Instruct-2412")
tokenizer = AutoTokenizer.from_pretrained("CarrotAI/Llama-3.2-Rabbit-Ko-3B-Instruct-2412")
```

## Score
### Performance Metrics
LogicKor
| Category | Single turn | Multi turn |
|---|---|---|
| ์ˆ˜ํ•™(Math) | 5.86 | 5.14 |
| ๋ฌธ๋ฒ•(Grammar) | 4.71 | 1.29 |
| ์ดํ•ด(Understanding) | 4.00 | 4.43 |
| ์ถ”๋ก (Reasoning) | 5.14 | 6.71 |
| ์ฝ”๋”ฉ(Coding) | 7.43 | 7.57 |
| ๊ธ€์“ฐ๊ธฐ(Writing) | 8.43 | 8.00 |
| **Total** | **5.93** | **5.52** |
| **Overall** | **5.73** | |


| Tasks  |Version|     Filter     |n-shot|        Metric         |   |Value |   |Stderr|
|--------|------:|----------------|-----:|-----------------------|---|-----:|---|------|
|gsm8k   |      3|flexible-extract|     5|exact_match            |โ†‘  |0.7013|ยฑ  |0.0126|
|        |       |strict-match    |     5|exact_match            |โ†‘  |0.2418|ยฑ  |0.0118|
|gsm8k-ko|      1|flexible-extract|     5|exact_match            |โ†‘  |0.4466|ยฑ  |0.0137|
|        |       |strict-match    |     5|exact_match            |โ†‘  |0.4420|ยฑ  |0.0137|
|ifeval  |      4|none            |     0|inst_level_loose_acc   |โ†‘  |0.8549|ยฑ  |   N/A|
|        |       |none            |     0|inst_level_strict_acc  |โ†‘  |0.8225|ยฑ  |   N/A|
|        |       |none            |     0|prompt_level_loose_acc |โ†‘  |0.7874|ยฑ  |0.0176|
|        |       |none            |     0|prompt_level_strict_acc|โ†‘  |0.7468|ยฑ  |0.0187|


|Task|Score|shot|
|---|---|---|
|haerae|43.26|5|

```
@article{Llama3.2RabbitKo3BInstruct,
  title={CarrotAI/Llama-3.2-Rabbit-Ko-3B-Instruct-2412 Card},
  author={CarrotAI (L, GEUN)},
  year={2024},
  url = {https://huggingface.co/CarrotAI/Llama-3.2-Rabbit-Ko-3B-Instruct-2412}
}
```