jujbob commited on
Commit
f612b63
β€’
1 Parent(s): ce5d4dd

Update README.md

Browse files
Files changed (1) hide show
  1. README.md +13 -165
README.md CHANGED
@@ -19,7 +19,7 @@ base_model:
19
  <!-- [CPU용 Colab μ–‘μžν™”λͺ¨λΈ μ½”λ“œμ˜ˆμ œ](https://colab.research.google.com/drive/129ZNVg5R2NPghUEFHKF0BRdxsZxinQcJ?usp=drive_link) -->
20
 
21
  ```bash
22
- 저희 Bllossom νŒ€μ—μ„œ llama3.1 기반의 ν•œκ΅­μ–΄-μ˜μ–΄ 이쀑 μ–Έμ–΄λͺ¨λΈ Bllossom-405B, Bllossom-8Bλ₯Ό κ³΅κ°œν•©λ‹ˆλ‹€.
23
  이번 Bllossom3.1-405BλŠ” preview λ²„μ „μœΌλ‘œ λ‹€μŒκ³Ό 같은 νŠΉμ§•μ„ λ³΄μž…λ‹ˆλ‹€.
24
  - Llama3.1-405B-Inst λŒ€λΉ„ 5~10% ν•œκ΅­μ–΄ μ„±λŠ₯이 ν–₯상 λ˜μ—ˆμŠ΅λ‹ˆλ‹€ (single turn κΈ°μ€€).
25
  - Llama3.1의 μ˜μ–΄ μ„±λŠ₯을 μ „ν˜€ μ†μƒμ‹œν‚€μ§€ μ•Šμ€ μ™„μ „ν•œ Bilingual λͺ¨λΈμž…λ‹ˆλ‹€.
@@ -28,22 +28,23 @@ base_model:
28
 
29
  ν•΄λ‹Ή λͺ¨λΈμ€ λ‹€μŒκ³Ό 같은 ν˜‘μ—…μ„ ν† λŒ€λ‘œ ꡬ좕 λ˜μ—ˆμŠ΅λ‹ˆλ‹€!
30
  - μ„œμšΈκ³ΌκΈ°λŒ€ MLPμ—°κ΅¬μ‹€μ˜ κ²½λŸ‰ν™” 사전 ν•™μŠ΅κΈ°λ²•μ΄ μ μš©λ˜μ—ˆμŠ΅λ‹ˆλ‹€.
 
31
  - HP의 computing 지원이 μžˆμ—ˆμŠ΅λ‹ˆλ‹€.
32
- - Common Crawl의 OscarνŒ€μ˜ 적극적인 데이터 지원이 μžˆμ—ˆμŠ΅λ‹ˆλ‹€
33
 
34
  μ–Έμ œλ‚˜ κ·Έλž¬λ“― ν•΄λ‹Ή λͺ¨λΈμ€ 상업적 이용이 κ°€λŠ₯ν•©λ‹ˆλ‹€. A100 6λŒ€λ§Œ μ€€λΉ„λ˜λ©΄ Bllossom을 μ΄μš©ν•΄ μ—¬λŸ¬λΆ„λ§Œμ˜ λͺ¨λΈμ„ λ§Œλ“€μ–΄λ³΄μ„Έμš” GPT4κ°€ 더이상 ν•„μš” μ—†μŠ΅λ‹ˆλ‹€.
35
  GPUμžμ›μ΄ λΆ€μ‘±ν•˜λ©΄ A100 3개 ν˜Ήμ€ A6000 4개둜 μ–‘μžν™” λͺ¨λΈμ„ μ΄μš©ν•΄ λ³΄μ„Έμš”. [μ–‘μžν™”λͺ¨λΈ](https://huggingface.co/MLP-KTLim/llama-3-Korean-Bllossom-8B-4bit)
36
 
37
  1. Bllossom-8BλŠ” μ„œμšΈκ³ΌκΈ°λŒ€, ν…Œλ””μΈ, μ—°μ„ΈλŒ€ μ–Έμ–΄μžμ› μ—°κ΅¬μ‹€μ˜ μ–Έμ–΄ν•™μžμ™€ ν˜‘μ—…ν•΄ λ§Œλ“  μ‹€μš©μ£Όμ˜κΈ°λ°˜ μ–Έμ–΄λͺ¨λΈμž…λ‹ˆλ‹€! μ•žμœΌλ‘œ 지속적인 μ—…λ°μ΄νŠΈλ₯Ό 톡해 κ΄€λ¦¬ν•˜κ² μŠ΅λ‹ˆλ‹€ 많이 ν™œμš©ν•΄μ£Όμ„Έμš” πŸ™‚
38
- 2. 초 κ°•λ ₯ν•œ Advanced-Bllossom 8B, 70Bλͺ¨λΈ, μ‹œκ°-μ–Έμ–΄λͺ¨λΈμ„ λ³΄μœ ν•˜κ³  μžˆμŠ΅λ‹ˆλ‹€! (κΆκΈˆν•˜μ‹ λΆ„μ€ κ°œλ³„ μ—°λ½μ£Όμ„Έμš”!!)
39
- 3. Bllossom은 NAACL2024, LREC-COLING2024 (ꡬ두) λ°œν‘œλ‘œ μ±„νƒλ˜μ—ˆμŠ΅λ‹ˆλ‹€.
40
  4. 쒋은 μ–Έμ–΄λͺ¨λΈ 계속 μ—…λ°μ΄νŠΈ ν•˜κ² μŠ΅λ‹ˆλ‹€!! ν•œκ΅­μ–΄ κ°•ν™”λ₯Όμœ„ν•΄ 곡동 μ—°κ΅¬ν•˜μ‹€λΆ„(νŠΉνžˆλ…Όλ¬Έ) μ–Έμ œλ“  ν™˜μ˜ν•©λ‹ˆλ‹€!!
41
- 특히 μ†ŒλŸ‰μ˜ GPU라도 λŒ€μ—¬ κ°€λŠ₯ν•œνŒ€μ€ μ–Έμ œλ“  μ—°λ½μ£Όμ„Έμš”! λ§Œλ“€κ³  싢은거 λ„μ™€λ“œλ €μš”.
42
  ```
43
 
44
  ```bash
45
  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:
46
- - Korean performance improved by 5-10% compared to Llama 3.1-405B-Inst (single turn basis).
47
  - A complete bilingual model that does not compromise the English performance of Llama 3.1.
48
  - Generates more natural and friendly Korean sentences compared to existing models.
49
  - Human evaluations and GPT evaluations (MT-Bench, LogicKor scoring 9, etc.) show performance similar to or slightly lower than GPT-4.
@@ -51,33 +52,6 @@ The Bllossom language model is a Korean-English bilingual language model based o
51
 
52
  **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)**
53
 
54
- ## Demo Video
55
-
56
- <div style="display: flex; justify-content: space-between;">
57
- <!-- 첫 번째 컬럼 -->
58
- <div style="width: 49%;">
59
- <a>
60
- <img src="https://github.com/lhsstn/lhsstn/blob/main/x-llava_dem.gif?raw=true" style="width: 100%; height: auto;">
61
- </a>
62
- <p style="text-align: center;">Bllossom-V Demo</p>
63
- </div>
64
-
65
- <!-- 두 번째 컬럼 (ν•„μš”ν•˜λ‹€λ©΄) -->
66
- <div style="width: 49%;">
67
- <a>
68
- <img src="https://github.com/lhsstn/lhsstn/blob/main/bllossom_demo_kakao.gif?raw=true" style="width: 70%; height: auto;">
69
- </a>
70
- <p style="text-align: center;">Bllossom Demo(Kakao)γ…€γ…€γ…€γ…€γ…€γ…€γ…€γ…€</p>
71
- </div>
72
- </div>
73
-
74
-
75
-
76
- # NEWS
77
- * [2024.06.18] We have reverted to the non-vocab-expansion model. However, we have significantly increased the amount of pre-training data to 250GB.
78
- * [2024.05.08] Vocab Expansion Model Update
79
- * [2024.04.25] We released Bllossom v2.0, based on llama-3
80
-
81
  ## Example code
82
 
83
  ### Colab Tutorial
@@ -93,7 +67,7 @@ pip install torch transformers==4.40.0 accelerate
93
  import transformers
94
  import torch
95
 
96
- model_id = "MLP-KTLim/llama-3-Korean-Bllossom-8B"
97
 
98
  pipeline = transformers.pipeline(
99
  "text-generation",
@@ -145,140 +119,14 @@ print(outputs[0]["generated_text"][len(prompt):])
145
  2. **뢁촌 ν•œμ˜₯λ§ˆμ„**
146
  - 전톡 ν•œοΏ½οΏ½μ΄ 잘 보쑴된 λ§ˆμ„λ‘œ, μ‘°μ„ μ‹œλŒ€μ˜ μƒν™œμƒμ„ λŠλ‚„ 수 μžˆμŠ΅λ‹ˆλ‹€.
147
 
148
- 3. **인사동**
149
- - 전톡 문화와 ν˜„λŒ€ 예술이 κ³΅μ‘΄ν•˜λŠ” 거리둜, λ‹€μ–‘ν•œ κ°€λŸ¬λ¦¬μ™€ 전톡 μŒμ‹μ μ΄ μžˆμŠ΅λ‹ˆλ‹€.
150
-
151
- 4. **μ²­κ³„μ²œ**
152
- - μ„œμšΈμ˜ 쀑심에 μœ„μΉ˜ν•œ 천문으둜, μ‘°κΉ…κ³Ό 산책을 즐길 수 μžˆλŠ” κ³³μž…λ‹ˆλ‹€.
153
-
154
- ### μ½”μŠ€ 2: μžμ—°κ³Ό μ‡Όν•‘
155
-
156
- 1. **남산 μ„œμšΈνƒ€μ›Œ**
157
- - μ„œμšΈμ˜ 전경을 ν•œλˆˆμ— λ³Ό 수 μžˆλŠ” 곳으둜, 특히 저녁 μ‹œκ°„λŒ€μ— 일λͺ°μ„ κ°μƒν•˜λŠ” 것이 μ’‹μŠ΅λ‹ˆλ‹€.
158
-
159
- 2. **λͺ…동**
160
- - μ‡Όν•‘κ³Ό μŒμ‹μ μ΄ μ¦λΉ„ν•œ μ§€μ—­μœΌλ‘œ, λ‹€μ–‘ν•œ λΈŒλžœλ“œμ™€ 전톡 μŒμ‹μ„ 맛볼 수 μžˆμŠ΅λ‹ˆλ‹€.
161
-
162
- 3. **ν•œκ°•κ³΅μ›**
163
- - μ„œμšΈμ˜ μ£Όμš” 곡원 쀑 ν•˜λ‚˜λ‘œ, μ‘°κΉ…, μžμ „κ±° 타기, λ°°λ‚­ 여행을 즐길 수 μžˆμŠ΅λ‹ˆλ‹€.
164
-
165
- 4. **ν™λŒ€**
166
- - μ Šμ€μ΄λ“€μ΄ 즐겨 μ°ΎλŠ” μ§€μ—­μœΌλ‘œ, λ‹€μ–‘ν•œ 카페, λ ˆμŠ€ν† λž‘, 클럽이 μžˆμŠ΅λ‹ˆλ‹€.
167
-
168
- ### μ½”μŠ€ 3: ν˜„λŒ€μ™€ μ „ν†΅μ˜ μ‘°ν™”
169
-
170
- 1. **λ™λŒ€λ¬Έ λ””μžμΈ ν”ŒλΌμž (DDP)**
171
- - ν˜„λŒ€μ μΈ κ±΄μΆ•λ¬Όλ‘œ, λ‹€μ–‘ν•œ μ „μ‹œμ™€ μ΄λ²€νŠΈκ°€ μ—΄λ¦¬λŠ” κ³³μž…λ‹ˆλ‹€.
172
-
173
- 2. **μ΄νƒœμ›**
174
- - λ‹€μ–‘ν•œ ꡭ제 μŒμ‹κ³Ό μΉ΄νŽ˜κ°€ μžˆλŠ” μ§€μ—­μœΌλ‘œ, λ‹€μ–‘ν•œ λ¬Έν™”λ₯Ό κ²½ν—˜ν•  수 μžˆμŠ΅λ‹ˆλ‹€.
175
-
176
- 3. **κ΄‘ν™”λ¬Έ**
177
- - μ„œμšΈμ˜ 쀑심에 μœ„μΉ˜ν•œ κ΄‘μž₯으둜, λ‹€μ–‘ν•œ 곡연과 행사가 μ—΄λ¦½λ‹ˆλ‹€.
178
-
179
- 4. **μ„œμšΈλžœλ“œ**
180
- - μ„œμšΈ 외곽에 μœ„μΉ˜ν•œ ν…Œλ§ˆνŒŒν¬λ‘œ, κ°€μ‘±λ‹¨μœ„ κ΄€κ΄‘κ°λ“€μ—κ²Œ 인기 μžˆλŠ” κ³³μž…λ‹ˆλ‹€.
181
-
182
- 이 μ½”μŠ€λ“€μ€ μ„œμšΈμ˜ λ‹€μ–‘ν•œ λ©΄λͺ¨λ₯Ό κ²½ν—˜ν•  수 μžˆλ„λ‘ κ΅¬μ„±λ˜μ–΄ μžˆμŠ΅λ‹ˆλ‹€. 각 μ½”μŠ€λ§ˆλ‹€ μ‹œκ°„μ„ μ‘°μ ˆν•˜κ³ , 개인의 관심사에 맞게 μ„ νƒν•˜μ—¬ λ°©λ¬Έν•˜λ©΄ 쒋을 것 κ°™μŠ΅λ‹ˆλ‹€. 즐거운 μ—¬ν–‰ λ˜μ„Έμš”!
183
- ```
184
-
185
- ### Python code with AutoModel
186
- ```python
187
-
188
- import os
189
- import torch
190
- from transformers import AutoTokenizer, AutoModelForCausalLM
191
-
192
- model_id = 'MLP-KTLim/llama-3-Korean-Bllossom-8B'
193
-
194
- tokenizer = AutoTokenizer.from_pretrained(model_id)
195
- model = AutoModelForCausalLM.from_pretrained(
196
- model_id,
197
- torch_dtype=torch.bfloat16,
198
- device_map="auto",
199
- )
200
-
201
- model.eval()
202
-
203
- PROMPT = '''You are a helpful AI assistant. Please answer the user's questions kindly. 당신은 유λŠ₯ν•œ AI μ–΄μ‹œμŠ€ν„΄νŠΈ μž…λ‹ˆλ‹€. μ‚¬μš©μžμ˜ μ§ˆλ¬Έμ— λŒ€ν•΄ μΉœμ ˆν•˜κ²Œ λ‹΅λ³€ν•΄μ£Όμ„Έμš”.'''
204
- instruction = "μ„œμšΈμ˜ 유λͺ…ν•œ κ΄€κ΄‘ μ½”μŠ€λ₯Ό λ§Œλ“€μ–΄μ€„λž˜?"
205
-
206
- messages = [
207
- {"role": "system", "content": f"{PROMPT}"},
208
- {"role": "user", "content": f"{instruction}"}
209
- ]
210
-
211
- input_ids = tokenizer.apply_chat_template(
212
- messages,
213
- add_generation_prompt=True,
214
- return_tensors="pt"
215
- ).to(model.device)
216
-
217
- terminators = [
218
- tokenizer.eos_token_id,
219
- tokenizer.convert_tokens_to_ids("<|eot_id|>")
220
- ]
221
-
222
- outputs = model.generate(
223
- input_ids,
224
- max_new_tokens=2048,
225
- eos_token_id=terminators,
226
- do_sample=True,
227
- temperature=0.6,
228
- top_p=0.9
229
- )
230
-
231
- print(tokenizer.decode(outputs[0][input_ids.shape[-1]:], skip_special_tokens=True))
232
- ```
233
- ```
234
- # 물둠이죠! μ„œμšΈμ€ λ‹€μ–‘ν•œ 문화와 역사, μžμ—°μ„ κ²ΈλΉ„ν•œ λ„μ‹œλ‘œ, λ§Žμ€ κ΄€κ΄‘ λͺ…μ†Œλ₯Ό μžλž‘ν•©λ‹ˆλ‹€. μ—¬κΈ° μ„œμšΈμ˜ 유λͺ…ν•œ κ΄€κ΄‘ μ½”μŠ€λ₯Ό μ†Œκ°œν•΄ λ“œλ¦΄κ²Œμš”.
235
-
236
- ### μ½”μŠ€ 1: 역사와 λ¬Έν™” 탐방
237
-
238
- 1. **경볡ꢁ**
239
- - μ„œμšΈμ˜ λŒ€ν‘œμ μΈ ꢁꢐ둜, μ‘°μ„  μ™•μ‘°μ˜ 역사와 λ¬Έν™”λ₯Ό μ²΄ν—˜ν•  수 μžˆλŠ” κ³³μž…λ‹ˆλ‹€.
240
-
241
- 2. **뢁촌 ν•œμ˜₯λ§ˆμ„**
242
- - 전톡 ν•œμ˜₯이 잘 보쑴된 λ§ˆμ„λ‘œ, μ‘°μ„ μ‹œλŒ€μ˜ μƒν™œμƒμ„ λŠλ‚„ 수 μžˆμŠ΅λ‹ˆλ‹€.
243
-
244
- 3. **인사동**
245
- - 전톡 문화와 ν˜„λŒ€ 예술이 κ³΅μ‘΄ν•˜λŠ” 거리둜, λ‹€μ–‘ν•œ κ°€λŸ¬λ¦¬μ™€ 전톡 μŒμ‹μ μ΄ μžˆμŠ΅λ‹ˆλ‹€.
246
-
247
- 4. **μ²­κ³„μ²œ**
248
- - μ„œμšΈμ˜ 쀑심에 μœ„μΉ˜ν•œ 천문으둜, μ‘°κΉ…κ³Ό 산책을 즐길 수 μžˆλŠ” κ³³μž…λ‹ˆλ‹€.
249
-
250
- ### μ½”μŠ€ 2: μžμ—°κ³Ό μ‡Όν•‘
251
-
252
- 1. **남산 μ„œμšΈνƒ€μ›Œ**
253
- - μ„œμšΈμ˜ 전경을 ν•œλˆˆμ— λ³Ό 수 μžˆλŠ” 곳으둜, 특히 저녁 μ‹œκ°„λŒ€μ— 일λͺ°μ„ κ°μƒν•˜λŠ” 것이 μ’‹μŠ΅λ‹ˆλ‹€.
254
-
255
- 2. **λͺ…동**
256
- - μ‡Όν•‘κ³Ό μŒμ‹μ μ΄ μ¦λΉ„ν•œ μ§€μ—­μœΌλ‘œ, λ‹€μ–‘ν•œ λΈŒλžœλ“œμ™€ 전톡 μŒμ‹μ„ 맛볼 수 μžˆμŠ΅λ‹ˆλ‹€.
257
-
258
- 3. **ν•œκ°•κ³΅μ›**
259
- - μ„œμšΈμ˜ μ£Όμš” 곡원 쀑 ν•˜λ‚˜λ‘œ, μ‘°κΉ…, μžμ „κ±° 타기, λ°°λ‚­ 여행을 즐길 수 μžˆμŠ΅λ‹ˆλ‹€.
260
-
261
- 4. **ν™λŒ€**
262
- - μ Šμ€μ΄λ“€μ΄ 즐겨 μ°ΎλŠ” μ§€μ—­μœΌλ‘œ, λ‹€μ–‘ν•œ 카페, λ ˆμŠ€ν† λž‘, 클럽이 μžˆμŠ΅λ‹ˆλ‹€.
263
-
264
- ### μ½”μŠ€ 3: ν˜„λŒ€μ™€ μ „ν†΅μ˜ μ‘°ν™”
265
-
266
- 1. **λ™λŒ€λ¬Έ λ””μžμΈ ν”ŒλΌμž (DDP)**
267
- - ν˜„λŒ€μ μΈ κ±΄μΆ•λ¬Όλ‘œ, λ‹€μ–‘ν•œ μ „μ‹œμ™€ μ΄λ²€νŠΈκ°€ μ—΄λ¦¬λŠ” κ³³μž…λ‹ˆλ‹€.
268
-
269
- 2. **μ΄νƒœμ›**
270
- - λ‹€μ–‘ν•œ ꡭ제 μŒμ‹κ³Ό μΉ΄νŽ˜κ°€ μžˆλŠ” μ§€μ—­μœΌλ‘œ, λ‹€μ–‘ν•œ λ¬Έν™”λ₯Ό κ²½ν—˜ν•  수 μžˆμŠ΅λ‹ˆλ‹€.
271
-
272
- 3. **κ΄‘ν™”λ¬Έ**
273
- - μ„œμšΈμ˜ 쀑심에 μœ„μΉ˜ν•œ κ΄‘μž₯으둜, λ‹€μ–‘ν•œ 곡연과 행사가 μ—΄λ¦½λ‹ˆλ‹€.
274
-
275
- 4. **μ„œμšΈλžœλ“œ**
276
- - μ„œμšΈ 외곽에 μœ„μΉ˜ν•œ ν…Œλ§ˆνŒŒν¬λ‘œ, κ°€μ‘±λ‹¨μœ„ κ΄€κ΄‘κ°λ“€μ—κ²Œ 인기 μžˆλŠ” κ³³μž…λ‹ˆλ‹€.
277
-
278
- 이 μ½”μŠ€λ“€μ€ μ„œμšΈμ˜ λ‹€μ–‘ν•œ λ©΄λͺ¨λ₯Ό κ²½ν—˜ν•  수 μžˆλ„λ‘ κ΅¬μ„±λ˜μ–΄ μžˆμŠ΅λ‹ˆλ‹€. 각 μ½”μŠ€λ§ˆλ‹€ μ‹œκ°„μ„ μ‘°μ ˆν•˜κ³ , 개인의 관심사에 맞게 οΏ½οΏ½οΏ½νƒν•˜μ—¬ λ°©λ¬Έν•˜λ©΄ 쒋을 것 κ°™μŠ΅λ‹ˆλ‹€. 즐거운 μ—¬ν–‰ λ˜μ„Έμš”!
279
  ```
280
 
 
281
 
 
 
 
282
 
283
  ## Citation
284
  **Language Model**
 
19
  <!-- [CPU용 Colab μ–‘μžν™”λͺ¨λΈ μ½”λ“œμ˜ˆμ œ](https://colab.research.google.com/drive/129ZNVg5R2NPghUEFHKF0BRdxsZxinQcJ?usp=drive_link) -->
20
 
21
  ```bash
22
+ 저희 Bllossom νŒ€μ—μ„œ llama3.1 기반의 ν•œκ΅­μ–΄-μ˜μ–΄ 이쀑 μ–Έμ–΄λͺ¨λΈ Bllossom-405Bλ₯Ό κ³΅κ°œν•©λ‹ˆλ‹€.
23
  이번 Bllossom3.1-405BλŠ” preview λ²„μ „μœΌλ‘œ λ‹€μŒκ³Ό 같은 νŠΉμ§•μ„ λ³΄μž…λ‹ˆλ‹€.
24
  - Llama3.1-405B-Inst λŒ€λΉ„ 5~10% ν•œκ΅­μ–΄ μ„±λŠ₯이 ν–₯상 λ˜μ—ˆμŠ΅λ‹ˆλ‹€ (single turn κΈ°μ€€).
25
  - Llama3.1의 μ˜μ–΄ μ„±λŠ₯을 μ „ν˜€ μ†μƒμ‹œν‚€μ§€ μ•Šμ€ μ™„μ „ν•œ Bilingual λͺ¨λΈμž…λ‹ˆλ‹€.
 
28
 
29
  ν•΄λ‹Ή λͺ¨λΈμ€ λ‹€μŒκ³Ό 같은 ν˜‘μ—…μ„ ν† λŒ€λ‘œ ꡬ좕 λ˜μ—ˆμŠ΅λ‹ˆλ‹€!
30
  - μ„œμšΈκ³ΌκΈ°λŒ€ MLPμ—°κ΅¬μ‹€μ˜ κ²½λŸ‰ν™” 사전 ν•™μŠ΅κΈ°λ²•μ΄ μ μš©λ˜μ—ˆμŠ΅λ‹ˆλ‹€.
31
+ - ν…Œλ””μΈμ˜ μ •κ΅ν•œ Instruction Tuningκ³Ό RAG 기술이 μ μš©λ˜μ—ˆμŠ΅λ‹ˆλ‹€.
32
  - HP의 computing 지원이 μžˆμ—ˆμŠ΅λ‹ˆλ‹€.
33
+ - Common Crawl μž¬λ‹¨μ˜ OscarνŒ€μ—μ„œ 적극적인 데이터 지원이 μžˆμ—ˆμŠ΅λ‹ˆλ‹€
34
 
35
  μ–Έμ œλ‚˜ κ·Έλž¬λ“― ν•΄λ‹Ή λͺ¨λΈμ€ 상업적 이용이 κ°€λŠ₯ν•©λ‹ˆλ‹€. A100 6λŒ€λ§Œ μ€€λΉ„λ˜λ©΄ Bllossom을 μ΄μš©ν•΄ μ—¬λŸ¬λΆ„λ§Œμ˜ λͺ¨λΈμ„ λ§Œλ“€μ–΄λ³΄μ„Έμš” GPT4κ°€ 더이상 ν•„μš” μ—†μŠ΅λ‹ˆλ‹€.
36
  GPUμžμ›μ΄ λΆ€μ‘±ν•˜λ©΄ A100 3개 ν˜Ήμ€ A6000 4개둜 μ–‘μžν™” λͺ¨λΈμ„ μ΄μš©ν•΄ λ³΄μ„Έμš”. [μ–‘μžν™”λͺ¨λΈ](https://huggingface.co/MLP-KTLim/llama-3-Korean-Bllossom-8B-4bit)
37
 
38
  1. Bllossom-8BλŠ” μ„œμšΈκ³ΌκΈ°λŒ€, ν…Œλ””μΈ, μ—°μ„ΈλŒ€ μ–Έμ–΄μžμ› μ—°κ΅¬μ‹€μ˜ μ–Έμ–΄ν•™μžμ™€ ν˜‘μ—…ν•΄ λ§Œλ“  μ‹€μš©μ£Όμ˜κΈ°λ°˜ μ–Έμ–΄λͺ¨λΈμž…λ‹ˆλ‹€! μ•žμœΌλ‘œ 지속적인 μ—…λ°μ΄νŠΈλ₯Ό 톡해 κ΄€λ¦¬ν•˜κ² μŠ΅λ‹ˆλ‹€ 많이 ν™œμš©ν•΄μ£Όμ„Έμš” πŸ™‚
39
+ 2. 초 κ°•λ ₯ν•œ Advanced-Bllossom λͺ¨λΈ, μ‹œκ°-μ–Έμ–΄ λͺ¨λΈμ„ λ³΄μœ ν•˜κ³  μžˆμŠ΅λ‹ˆλ‹€! (κΆκΈˆν•˜μ‹ λΆ„μ€ κ°œλ³„ μ—°λ½μ£Όμ„Έμš”!!)
40
+ 3. Bllossom은 NAACL2024, LREC-COLING2024 (ꡬ두) λ°œν‘œλ˜μ—ˆμŠ΅λ‹ˆλ‹€.
41
  4. 쒋은 μ–Έμ–΄λͺ¨λΈ 계속 μ—…λ°μ΄νŠΈ ν•˜κ² μŠ΅λ‹ˆλ‹€!! ν•œκ΅­μ–΄ κ°•ν™”λ₯Όμœ„ν•΄ 곡동 μ—°κ΅¬ν•˜μ‹€λΆ„(νŠΉνžˆλ…Όλ¬Έ) μ–Έμ œλ“  ν™˜μ˜ν•©λ‹ˆλ‹€!!
42
+ 그리고 μ†ŒλŸ‰μ˜ GPU라도 λŒ€μ—¬ κ°€λŠ₯ν•œνŒ€μ€ μ–Έμ œλ“  μ—°λ½μ£Όμ„Έμš”! λ§Œλ“€κ³  싢은거 λ„μ™€λ“œλ €μš”.
43
  ```
44
 
45
  ```bash
46
  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:
47
+ - Korean performance improved by 5-10% compared to Llama 3.1-405B-Inst (on Single Turn Eval).
48
  - A complete bilingual model that does not compromise the English performance of Llama 3.1.
49
  - Generates more natural and friendly Korean sentences compared to existing models.
50
  - Human evaluations and GPT evaluations (MT-Bench, LogicKor scoring 9, etc.) show performance similar to or slightly lower than GPT-4.
 
52
 
53
  **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)**
54
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
55
  ## Example code
56
 
57
  ### Colab Tutorial
 
67
  import transformers
68
  import torch
69
 
70
+ model_id = "MLP-KTLim/llama-3.1-Korean-Bllossom-405B"
71
 
72
  pipeline = transformers.pipeline(
73
  "text-generation",
 
119
  2. **뢁촌 ν•œμ˜₯λ§ˆμ„**
120
  - 전톡 ν•œοΏ½οΏ½μ΄ 잘 보쑴된 λ§ˆμ„λ‘œ, μ‘°μ„ μ‹œλŒ€μ˜ μƒν™œμƒμ„ λŠλ‚„ 수 μžˆμŠ΅λ‹ˆλ‹€.
121
 
122
+ ...
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
123
  ```
124
 
125
+ ## Supported by
126
 
127
+ - Hewlett Packard (HP) Enterprise
128
+ - Common Crawl
129
+ - AICA
130
 
131
  ## Citation
132
  **Language Model**