File size: 9,107 Bytes
b5e840c
398e261
 
1e8ebc7
 
 
b5e840c
398e261
b5e840c
 
1e8ebc7
 
 
 
 
 
 
 
8144d8a
1e8ebc7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
79718e6
1e8ebc7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7b571c5
1e8ebc7
 
 
 
 
 
 
 
 
030ad7e
1e8ebc7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
7b571c5
1e8ebc7
 
 
 
 
 
 
 
 
 
030ad7e
1e8ebc7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
49099bf
1e8ebc7
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
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
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
---
base_model:
- meta-llama/Meta-Llama-3-70B
language:
- en
- ko
library_name: transformers
license: llama3
---

<a href="https://github.com/MLP-Lab/Bllossom">
  <img src="https://github.com/teddysum/bllossom/blob/main//bllossom_icon.png?raw=true" width="40%" height="50%">
</a>

# Bllossom | [Demo]() | [Homepage](https://www.bllossom.ai/) | [Github](https://github.com/MLP-Lab/Bllossom) | [Colab-tutorial](https://colab.research.google.com/drive/1fBOzUVZ6NRKk_ugeoTbAOokWKqSN47IG?usp=sharing) |


```bash
์ €ํฌ Bllossom ํ”„๋กœ์ ํŠธ ํŒ€์—์„œ ํ•œ๊ตญ์–ด-์˜์–ด ์ด์ค‘ ์–ธ์–ด๋ชจ๋ธ์ธ Bllossom-70.8B๋ฅผ ๊ณต๊ฐœํ–ˆ์Šต๋‹ˆ๋‹ค!
์„œ์šธ๊ณผ๊ธฐ๋Œ€ ์Šˆํผ์ปดํ“จํŒ… ์„ผํ„ฐ์˜ ์ง€์›์œผ๋กœ 100GB๊ฐ€๋„˜๋Š” ํ•œ๊ตญ์–ด๋กœ ๋ชจ๋ธ์ „์ฒด๋ฅผ ํ’€ํŠœ๋‹ํ•œ ํ•œ๊ตญ์–ด ๊ฐ•ํ™” ์ด์ค‘์–ธ์–ด ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค!
ํ•œ๊ตญ์–ด ์ž˜ํ•˜๋Š” ๋ชจ๋ธ ์ฐพ๊ณ  ์žˆ์ง€ ์•Š์œผ์…จ๋‚˜์š”?
 - ํ•œ๊ตญ์–ด ์ตœ์ดˆ! ๋ฌด๋ ค 3๋งŒ๊ฐœ๊ฐ€ ๋„˜๋Š” ํ•œ๊ตญ์–ด ์–ดํœ˜ํ™•์žฅ
 - Llama3๋Œ€๋น„ ๋Œ€๋žต 25% ๋” ๊ธด ๊ธธ์ด์˜ ํ•œ๊ตญ์–ด Context ์ฒ˜๋ฆฌ๊ฐ€๋Šฅ
 - ํ•œ๊ตญ์–ด-์˜์–ด Pararell Corpus๋ฅผ ํ™œ์šฉํ•œ ํ•œ๊ตญ์–ด-์˜์–ด ์ง€์‹์—ฐ๊ฒฐ (์‚ฌ์ „ํ•™์Šต)
 - ํ•œ๊ตญ์–ด ๋ฌธํ™”, ์–ธ์–ด๋ฅผ ๊ณ ๋ คํ•ด ์–ธ์–ดํ•™์ž๊ฐ€ ์ œ์ž‘ํ•œ ๋ฐ์ดํ„ฐ๋ฅผ ํ™œ์šฉํ•œ ๋ฏธ์„ธ์กฐ์ •
 - ๊ฐ•ํ™”ํ•™์Šต
์ด ๋ชจ๋“ ๊ฒŒ ํ•œ๊บผ๋ฒˆ์— ์ ์šฉ๋˜๊ณ  ์ƒ์—…์  ์ด์šฉ์ด ๊ฐ€๋Šฅํ•œ Bllossom์„ ์ด์šฉํ•ด ์—ฌ๋Ÿฌ๋ถ„ ๋งŒ์˜ ๋ชจ๋ธ์„ ๋งŒ๋“ค์–ด๋ณด์„ธ์šฅ!
GPU๊ฐ€ ๋ถ€์กฑํ•˜๋ฉด ์–‘์žํ™” ๋ชจ๋ธ๋กœ ๋ฐ”๋กœ ์„œ๋น„์Šค๋ฅผ ํ™œ์šฉํ•ด ๋ณด์„ธ์š” [์–‘์žํ™”๋ชจ๋ธ](https://huggingface.co/Bllossom/llama-3-Korean-Bllossom-70B-gguf-Q4_K_M)!!

1. Bllossom-70.8B๋Š” ์„œ์šธ๊ณผ๊ธฐ๋Œ€, ํ…Œ๋””์ธ, ์—ฐ์„ธ๋Œ€ ์–ธ์–ด์ž์› ์—ฐ๊ตฌ์‹ค์˜ ์–ธ์–ดํ•™์ž์™€ ํ˜‘์—…ํ•ด ๋งŒ๋“  ์‹ค์šฉ์ฃผ์˜๊ธฐ๋ฐ˜ ์–ธ์–ด๋ชจ๋ธ์ž…๋‹ˆ๋‹ค! ์•ž์œผ๋กœ ์ง€์†์ ์ธ ์—…๋ฐ์ดํŠธ๋ฅผ ํ†ตํ•ด ๊ด€๋ฆฌํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค ๋งŽ์ด ํ™œ์šฉํ•ด์ฃผ์„ธ์š” ๐Ÿ™‚
2. ์ดˆ ๊ฐ•๋ ฅํ•œ Advanced-Bllossom 8B, 70B๋ชจ๋ธ, ์‹œ๊ฐ-์–ธ์–ด๋ชจ๋ธ์„ ๋ณด์œ ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค! (๊ถ๊ธˆํ•˜์‹ ๋ถ„์€ ๊ฐœ๋ณ„ ์—ฐ๋ฝ์ฃผ์„ธ์š”!!)
3. Bllossom์€ NAACL2024, LREC-COLING2024 (๊ตฌ๋‘) ๋ฐœํ‘œ๋กœ ์ฑ„ํƒ๋˜์—ˆ์Šต๋‹ˆ๋‹ค.
4. ์ข‹์€ ์–ธ์–ด๋ชจ๋ธ ๊ณ„์† ์—…๋ฐ์ดํŠธ ํ•˜๊ฒ ์Šต๋‹ˆ๋‹ค!! ํ•œ๊ตญ์–ด ๊ฐ•ํ™”๋ฅผ์œ„ํ•ด ๊ณต๋™ ์—ฐ๊ตฌํ•˜์‹ค๋ถ„(ํŠนํžˆ๋…ผ๋ฌธ) ์–ธ์ œ๋“  ํ™˜์˜ํ•ฉ๋‹ˆ๋‹ค!! 
   ํŠนํžˆ ์†Œ๋Ÿ‰์˜ GPU๋ผ๋„ ๋Œ€์—ฌ ๊ฐ€๋Šฅํ•œํŒ€์€ ์–ธ์ œ๋“  ์—ฐ๋ฝ์ฃผ์„ธ์š”! ๋งŒ๋“ค๊ณ  ์‹ถ์€๊ฑฐ ๋„์™€๋“œ๋ ค์š”.
```

The Bllossom language model is a Korean-English bilingual language model based on the open-source LLama3. It enhances the connection of knowledge between Korean and English. It has the following features:

* **Knowledge Linking**: Linking Korean and English knowledge through additional training
* **Vocabulary Expansion**: Expansion of Korean vocabulary to enhance Korean expressiveness.
* **Instruction Tuning**: Tuning using custom-made instruction following data specialized for Korean language and Korean culture
* **Human Feedback**: DPO has been applied
* **Vision-Language Alignment**: Aligning the vision transformer with this language model 

**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)**

## Demo Video

<div style="display: flex; justify-content: space-between;">
  <!-- ์ฒซ ๋ฒˆ์งธ ์ปฌ๋Ÿผ -->
  <div style="width: 49%;">
    <a>
      <img src="https://github.com/lhsstn/lhsstn/blob/main/x-llava_dem.gif?raw=true" style="width: 100%; height: auto;">
    </a>
    <p style="text-align: center;">Bllossom-V Demo</p>
  </div>

  <!-- ๋‘ ๋ฒˆ์งธ ์ปฌ๋Ÿผ (ํ•„์š”ํ•˜๋‹ค๋ฉด) -->
  <div style="width: 49%;">
    <a>
      <img src="https://github.com/lhsstn/lhsstn/blob/main/bllossom_demo_kakao.gif?raw=true" style="width: 70%; height: auto;">
    </a>
    <p style="text-align: center;">Bllossom Demo(Kakao)ใ…คใ…คใ…คใ…คใ…คใ…คใ…คใ…ค</p>
  </div>
</div>



## NEWS
* [2024.08.30] ์‚ฌ์ „ํ•™์Šต๋Ÿ‰์„ 250GB๊นŒ์ง€ ๋Š˜๋ฆฐ Bllossom ELO๋ชจ๋ธ๋กœ ์—…๋ฐ์ดํŠธ ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ๋‹ค๋งŒ ๋‹จ์–ดํ™•์žฅ์€ ํ•˜์ง€ ์•Š์•˜์Šต๋‹ˆ๋‹ค. ๊ธฐ์กด ๋‹จ์–ดํ™•์žฅ๋œ long-context ๋ชจ๋ธ์„ ํ™œ์šฉํ•˜๊ณ  ์‹ถ์œผ์‹ ๋ถ„์€ ๊ฐœ์ธ์—ฐ๋ฝ์ฃผ์„ธ์š”!
* [2024.05.08] Vocab Expansion Model Update
* [2024.04.25] We released Bllossom v2.0, based on llama-3
* [2023/12] We released Bllossom-Vision v1.0, based on Bllossom
* [2023/08] We released Bllossom v1.0, based on llama-2. 
* [2023/07] We released Bllossom v0.7, based on polyglot-ko.


## 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-Korean-Bllossom-70B"

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‹ ์€ ์œ ๋Šฅํ•œ AI ์–ด์‹œ์Šคํ„ดํŠธ ์ž…๋‹ˆ๋‹ค. ์‚ฌ์šฉ์ž์˜ ์งˆ๋ฌธ์— ๋Œ€ํ•ด ์นœ์ ˆํ•˜๊ฒŒ ๋‹ต๋ณ€ํ•ด์ฃผ์„ธ์š”.'''
instruction = "์„œ์šธ๊ณผํ•™๊ธฐ์ˆ ๋Œ€ํ•™๊ต MLP์—ฐ๊ตฌ์‹ค์— ๋Œ€ํ•ด ์†Œ๊ฐœํ•ด์ค˜"

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):])

# ์„œ์šธ๊ณผํ•™๊ธฐ์ˆ ๋Œ€ํ•™๊ต MLP์—ฐ๊ตฌ์‹ค์€ ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ ์ž์—ฐ์–ด์ฒ˜๋ฆฌ ์—ฐ๊ตฌ๋ฅผ ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๊ตฌ์„ฑ์›์€ ์ž„๊ฒฝํƒœ ๊ต์ˆ˜์™€ ๊น€๋ฏผ์ค€, ๊น€์ƒ๋ฏผ, ์ตœ์ฐฝ์ˆ˜, ์›์ธํ˜ธ, ์œ ํ•œ๊ฒฐ, ์ž„ํ˜„์„, ์†ก์Šน์šฐ, ์œก์ •ํ›ˆ, ์‹ ๋™์žฌ ํ•™์ƒ์ด ์žˆ์Šต๋‹ˆ๋‹ค.
```

### Python code with AutoModel
```python

import os
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM

model_id = 'Bllossom/llama-3-Korean-Bllossom-70B'

tokenizer = AutoTokenizer.from_pretrained(model_id)
model = AutoModelForCausalLM.from_pretrained(
    model_id,
    torch_dtype=torch.bfloat16,
    device_map="auto",
)

model.eval()

PROMPT = '''You are a helpful AI assistant. Please answer the user's questions kindly. ๋‹น์‹ ์€ ์œ ๋Šฅํ•œ AI ์–ด์‹œ์Šคํ„ดํŠธ ์ž…๋‹ˆ๋‹ค. ์‚ฌ์šฉ์ž์˜ ์งˆ๋ฌธ์— ๋Œ€ํ•ด ์นœ์ ˆํ•˜๊ฒŒ ๋‹ต๋ณ€ํ•ด์ฃผ์„ธ์š”.'''
instruction = "์„œ์šธ๊ณผํ•™๊ธฐ์ˆ ๋Œ€ํ•™๊ต MLP์—ฐ๊ตฌ์‹ค์— ๋Œ€ํ•ด ์†Œ๊ฐœํ•ด์ค˜"

messages = [
    {"role": "system", "content": f"{PROMPT}"},
    {"role": "user", "content": f"{instruction}"}
    ]

input_ids = tokenizer.apply_chat_template(
    messages,
    add_generation_prompt=True,
    return_tensors="pt"
).to(model.device)

terminators = [
    tokenizer.eos_token_id,
    tokenizer.convert_tokens_to_ids("<|eot_id|>")
]

outputs = model.generate(
    input_ids,
    max_new_tokens=2048,
    eos_token_id=terminators,
    do_sample=True,
    temperature=0.6,
    top_p=0.9
)

print(tokenizer.decode(outputs[0][input_ids.shape[-1]:], skip_special_tokens=True))
# ์„œ์šธ๊ณผํ•™๊ธฐ์ˆ ๋Œ€ํ•™๊ต MLP์—ฐ๊ตฌ์‹ค์€ ๋ฉ€ํ‹ฐ๋ชจ๋‹ฌ ์ž์—ฐ์–ด์ฒ˜๋ฆฌ ์—ฐ๊ตฌ๋ฅผ ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๊ตฌ์„ฑ์›์€ ์ž„๊ฒฝํƒœ ๊ต์ˆ˜์™€ ๊น€๋ฏผ์ค€, ๊น€์ƒ๋ฏผ, ์ตœ์ฐฝ์ˆ˜, ์›์ธํ˜ธ, ์œ ํ•œ๊ฒฐ, ์ž„ํ˜„์„, ์†ก์Šน์šฐ, ์œก์ •ํ›ˆ, ์‹ ๋™์žฌ ํ•™์ƒ์ด ์žˆ์Šต๋‹ˆ๋‹ค.
```



## 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
 - ์ž„๊ฒฝํƒœ(KyungTae Lim), Professor at Seoultech. `[email protected]`
 - ํ•จ์˜๊ท (Younggyun Hahm), CEO of Teddysum. `[email protected]`
 - ๊น€ํ•œ์ƒ˜(Hansaem Kim), Professor at Yonsei. `[email protected]`

## Contributor
 - ์ตœ์ฐฝ์ˆ˜(Chansu Choi), [email protected]
 - ๊น€์ƒ๋ฏผ(Sangmin Kim), [email protected]
 - ์›์ธํ˜ธ(Inho Won), [email protected]
 - ๊น€๋ฏผ์ค€(Minjun Kim), [email protected] 
 - ์†ก์Šน์šฐ(Seungwoo Song), [email protected]
 - ์‹ ๋™์žฌ(Dongjae Shin), [email protected]
 - ์ž„ํ˜„์„(Hyeonseok Lim), [email protected]
 - ์œก์ •ํ›ˆ(Jeonghun Yuk), [email protected]
 - ์œ ํ•œ๊ฒฐ(Hangyeol Yoo), [email protected]
 - ์†ก์„œํ˜„(Seohyun Song), [email protected]