File size: 7,780 Bytes
ce5d4dd aa12552 ce5d4dd b30ec8d ce5d4dd f612b63 ce5d4dd f612b63 ce5d4dd f612b63 ce5d4dd b30ec8d ce5d4dd b30ec8d f612b63 ce5d4dd f612b63 ce5d4dd f612b63 ce5d4dd 3185b9c ce5d4dd f612b63 ce5d4dd f612b63 ce5d4dd aa12552 f612b63 ce5d4dd |
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 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 |
---
language:
- en
- ko
license: llama3.1
library_name: transformers
base_model:
- meta-llama/Meta-Llama-3.1-405B
---
<a href="https://github.com/MLP-Lab/Bllossom">
<img src="https://github.com/teddysum/bllossom/blob/main//bllossom_icon.png?raw=true" width="30%" height="30%">
</a>
# Update!
* [2024.08.08] preview ๋ชจ๋ธ์ด ์ต์ด ์
๋ฐ์ดํธ ๋์์ต๋๋ค. A100 120๋ ๊ท๋ชจ์ ์ปดํจํ
ํ์๋ก ํ์ต ์งํ์ค์ผ๋ก ๋ชจ๋ธ์ ๊ณ์ ์
๋ฐ์ดํธ๋ ์์ ์
๋๋ค.
# Bllossom | [Demo]() | [Homepage](https://www.bllossom.ai/) | [Github](https://github.com/MLP-Lab/Bllossom) |
<!-- [GPU์ฉ Colab ์ฝ๋์์ ](https://colab.research.google.com/drive/1fBOzUVZ6NRKk_ugeoTbAOokWKqSN47IG?usp=sharing) | -->
<!-- [CPU์ฉ Colab ์์ํ๋ชจ๋ธ ์ฝ๋์์ ](https://colab.research.google.com/drive/129ZNVg5R2NPghUEFHKF0BRdxsZxinQcJ?usp=drive_link) -->
```bash
์ ํฌ Bllossom ํ์์ llama3.1 ๊ธฐ๋ฐ์ ํ๊ตญ์ด-์์ด ์ด์ค ์ธ์ด๋ชจ๋ธ Bllossom-405B๋ฅผ ๊ณต๊ฐํฉ๋๋ค.
์ด๋ฒ Bllossom3.1-405B๋ preview ๋ฒ์ ์ผ๋ก ๋ค์๊ณผ ๊ฐ์ ํน์ง์ ๋ณด์
๋๋ค.
- Llama3.1-405B-Inst ๋๋น 5~10% ํ๊ตญ์ด ์ฑ๋ฅ์ด ํฅ์ ๋์์ต๋๋ค (single turn ๊ธฐ์ค).
- Llama3.1์ ์์ด ์ฑ๋ฅ์ ์ ํ ์์์ํค์ง ์์ ์์ ํ Bilingual ๋ชจ๋ธ์
๋๋ค.
- ๊ธฐ์กด ๋ชจ๋ธ ๋๋น ์์ฐ์ค๋ฝ๊ณ ์น์ ํ ํ๊ตญ์ด ๋ฌธ์ฅ์ ์์ฑํฉ๋๋ค.
- ์ธ๊ฐํ๊ฐ, GPTํ๊ฐ(MT-Bench, LogicKor 9์ ๋ฑ) ๊ฒฐ๊ณผ GPT4์ ์ ์ฌํ๊ฑฐ๋ ์ฝ๊ฐ ๋ฎ์ ์ฑ๋ฅ์ ๋ณด์ฌ์ค๋๋ค.
ํด๋น ๋ชจ๋ธ์ ๋ค์๊ณผ ๊ฐ์ ํ์
์ ํ ๋๋ก ๊ตฌ์ถ ๋์์ต๋๋ค!
- ์์ธ๊ณผ๊ธฐ๋ MLP์ฐ๊ตฌ์ค์ ๊ฒฝ๋ํ ์ฌ์ ํ์ต๊ธฐ๋ฒ์ด ์ ์ฉ๋์์ต๋๋ค.
- ํ
๋์ธ์ ์ ๊ตํ Instruction Tuning๊ณผ RAG ๊ธฐ์ ์ด ์ ์ฉ๋์์ต๋๋ค.
- HP์ computing ์ง์์ด ์์์ต๋๋ค.
- Common Crawl ์ฌ๋จ์ Oscarํ์์ ์ ๊ทน์ ์ธ ๋ฐ์ดํฐ ์ง์์ด ์์์ต๋๋ค
์ธ์ ๋ ๊ทธ๋ฌ๋ฏ ํด๋น ๋ชจ๋ธ์ ์์
์ ์ด์ฉ์ด ๊ฐ๋ฅํฉ๋๋ค. A100 6๋๋ง ์ค๋น๋๋ฉด Bllossom์ ์ด์ฉํด ์ฌ๋ฌ๋ถ๋ง์ ๋ชจ๋ธ์ ๋ง๋ค์ด๋ณด์ธ์ GPT4๊ฐ ๋์ด์ ํ์ ์์ต๋๋ค.
GPU์์์ด ๋ถ์กฑํ๋ฉด A100 3๊ฐ ํน์ A6000 4๊ฐ๋ก ์์ํ ๋ชจ๋ธ์ ์ด์ฉํด ๋ณด์ธ์. [์์ํ๋ชจ๋ธ](https://huggingface.co./MLP-KTLim/llama-3.1-Korean-Bllossom-405B-gguf-Q4_K_M)
1. Bllossom-8B๋ ์์ธ๊ณผ๊ธฐ๋, ํ
๋์ธ, ์ฐ์ธ๋ ์ธ์ด์์ ์ฐ๊ตฌ์ค์ ์ธ์ดํ์์ ํ์
ํด ๋ง๋ ์ค์ฉ์ฃผ์๊ธฐ๋ฐ ๋ฌด๋ฃ ์ธ์ด๋ชจ๋ธ๋ก 2023๋
๋ถํฐ ์ง์์ ์ธ ์
๋ฐ์ดํธ๋ฅผ ํตํด ๊ด๋ฆฌํด ์ค๊ณ ์์ต๋๋ค. ๋ง์ด ํ์ฉํด์ฃผ์ธ์ ๐
2. ์ด ๊ฐ๋ ฅํ Advanced-Bllossom ๋ชจ๋ธ, ์๊ฐ-์ธ์ด ๋ชจ๋ธ์ ๋ณด์ ํ๊ณ ์์ต๋๋ค! (๊ถ๊ธํ์ ๋ถ์ ๊ฐ๋ณ ์ฐ๋ฝ์ฃผ์ธ์!!)
3. Bllossom์ NAACL2024, LREC-COLING2024 (๊ตฌ๋) ๋ฐํ๋์์ต๋๋ค.
4. ์ข์ ์ธ์ด๋ชจ๋ธ ๊ณ์ ์
๋ฐ์ดํธ ํ๊ฒ ์ต๋๋ค!! ํ๊ตญ์ด ๊ฐํ๋ฅผ์ํด ๊ณต๋ ์ฐ๊ตฌํ์ค๋ถ(ํนํ๋
ผ๋ฌธ) ์ธ์ ๋ ํ์ํฉ๋๋ค!!
๊ทธ๋ฆฌ๊ณ ์๋์ GPU๋ผ๋ ๋์ฌ ๊ฐ๋ฅํํ์ ์ธ์ ๋ ์ฐ๋ฝ์ฃผ์ธ์! ๋ง๋ค๊ณ ์ถ์๊ฑฐ ๋์๋๋ ค์.
```
```bash
The Bllossom language model is a Korean-English bilingual language model based on the open-source LLama3.1. It enhances the connection of knowledge between Korean and English. It has the following features:
- Korean performance improved by 5-10% compared to Llama 3.1-405B-Inst (on Single Turn Eval).
- A complete bilingual model that does not compromise the English performance of Llama 3.1.
- Generates more natural and friendly Korean sentences compared to existing models.
- Human evaluations and GPT evaluations (MT-Bench, LogicKor scoring 9, etc.) show performance similar to or slightly lower than GPT-4.
```
**This model developed by [MLPLab at Seoultech](http://mlp.seoultech.ac.kr), [Teddysum](http://teddysum.ai/) and [Yonsei Univ](https://sites.google.com/view/hansaemkim/hansaem-kim)**
## Example code
### Colab Tutorial
- [Inference-Code-Link](https://colab.research.google.com/drive/1fBOzUVZ6NRKk_ugeoTbAOokWKqSN47IG?usp=sharing)
### Install Dependencies
```bash
pip install torch transformers==4.40.0 accelerate
```
### Python code with Pipeline
```python
import transformers
import torch
model_id = "Bllossom/llama-3.1-Korean-Bllossom-405B"
pipeline = transformers.pipeline(
"text-generation",
model=model_id,
model_kwargs={"torch_dtype": torch.bfloat16},
device_map="auto",
)
pipeline.model.eval()
PROMPT = '''You are a helpful AI assistant. Please answer the user's questions kindly. ๋น์ ์ ์ ๋ฅํ AI ์ด์์คํดํธ ์
๋๋ค. ์ฌ์ฉ์์ ์ง๋ฌธ์ ๋ํด ์น์ ํ๊ฒ ๋ต๋ณํด์ฃผ์ธ์.'''
instruction = "์์ธ์ ์ ๋ช
ํ ๊ด๊ด ์ฝ์ค๋ฅผ ๋ง๋ค์ด์ค๋?"
messages = [
{"role": "system", "content": f"{PROMPT}"},
{"role": "user", "content": f"{instruction}"}
]
prompt = pipeline.tokenizer.apply_chat_template(
messages,
tokenize=False,
add_generation_prompt=True
)
terminators = [
pipeline.tokenizer.eos_token_id,
pipeline.tokenizer.convert_tokens_to_ids("<|eot_id|>")
]
outputs = pipeline(
prompt,
max_new_tokens=2048,
eos_token_id=terminators,
do_sample=True,
temperature=0.6,
top_p=0.9
)
print(outputs[0]["generated_text"][len(prompt):])
```
```
# ๋ฌผ๋ก ์ด์ฃ ! ์์ธ์ ๋ค์ํ ๋ฌธํ์ ์ญ์ฌ, ์์ฐ์ ๊ฒธ๋นํ ๋์๋ก, ๋ง์ ๊ด๊ด ๋ช
์๋ฅผ ์๋ํฉ๋๋ค. ์ฌ๊ธฐ ์์ธ์ ์ ๋ช
ํ ๊ด๊ด ์ฝ์ค๋ฅผ ์๊ฐํด ๋๋ฆด๊ฒ์.
### ์ฝ์ค 1: ์ญ์ฌ์ ๋ฌธํ ํ๋ฐฉ
1. **๊ฒฝ๋ณต๊ถ**
- ์์ธ์ ๋ํ์ ์ธ ๊ถ๊ถ๋ก, ์กฐ์ ์์กฐ์ ์ญ์ฌ์ ๋ฌธํ๋ฅผ ์ฒดํํ ์ ์๋ ๊ณณ์
๋๋ค.
2. **๋ถ์ด ํ์ฅ๋ง์**
- ์ ํต ํ์ฅ์ด ์ ๋ณด์กด๋ ๋ง์๋ก, ์กฐ์ ์๋์ ์ํ์์ ๋๋ ์ ์์ต๋๋ค.
...
```
## Supported by
- Hewlett Packard (HP) Enterprise <img src="https://upload.wikimedia.org/wikipedia/commons/thumb/4/46/Hewlett_Packard_Enterprise_logo.svg/2880px-Hewlett_Packard_Enterprise_logo.svg.png" width="20%" height="20%">
- Common Crawl <img src="https://cdn.prod.website-files.com/6479b8d98bf5dcb4a69c4f31/649b5869af56f6df617cfb1f_CC_Logo_Blue_Auto.svg" width="20%" height="20%">
- AICA
## Citation
**Language Model**
```text
@misc{bllossom,
author = {ChangSu Choi, Yongbin Jeong, Seoyoon Park, InHo Won, HyeonSeok Lim, SangMin Kim, Yejee Kang, Chanhyuk Yoon, Jaewan Park, Yiseul Lee, HyeJin Lee, Younggyun Hahm, Hansaem Kim, KyungTae Lim},
title = {Optimizing Language Augmentation for Multilingual Large Language Models: A Case Study on Korean},
year = {2024},
journal = {LREC-COLING 2024},
paperLink = {\url{https://arxiv.org/pdf/2403.10882}},
},
}
```
**Vision-Language Model**
```text
@misc{bllossom-V,
author = {Dongjae Shin, Hyunseok Lim, Inho Won, Changsu Choi, Minjun Kim, Seungwoo Song, Hangyeol Yoo, Sangmin Kim, Kyungtae Lim},
title = {X-LLaVA: Optimizing Bilingual Large Vision-Language Alignment},
year = {2024},
publisher = {GitHub},
journal = {NAACL 2024 findings},
paperLink = {\url{https://arxiv.org/pdf/2403.11399}},
},
}
```
## Contact
- ์๊ฒฝํ(KyungTae Lim), Professor at Seoultech. `[email protected]`
- ํจ์๊ท (Younggyun Hahm), CEO of Teddysum. `[email protected]`
- ๊นํ์(Hansaem Kim), Professor at Yonsei. `[email protected]`
## Contributor
- ์ต์ฐฝ์(Chansu Choi), [email protected]
- ๊น์๋ฏผ(Sangmin Kim), [email protected]
- ์์ธํธ(Inho Won), [email protected]
- ๊น๋ฏผ์ค(Minjun Kim), [email protected]
- ์ก์น์ฐ(Seungwoo Song), [email protected]
- ์ ๋์ฌ(Dongjae Shin), [email protected]
- ์ํ์(Hyeonseok Lim), [email protected]
- ์ก์ ํ(Jeonghun Yuk), [email protected]
- ์ ํ๊ฒฐ(Hangyeol Yoo), [email protected]
- ์ก์ํ(Seohyun Song), [email protected] |