KIST-robot-intelligence commited on
Commit
58316b8
โ€ข
1 Parent(s): 06d9011

Create README.md

Browse files
Files changed (1) hide show
  1. README.md +133 -0
README.md ADDED
@@ -0,0 +1,133 @@
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
+ ---
2
+ language: ko
3
+ pipeline_tag: text-generation
4
+ license: llama3
5
+ ---
6
+
7
+
8
+
9
+ #KIST-robot-intelligence/
10
+ KONI-Llama3-8B-Instruct-20240729-GGUF-Quantization
11
+
12
+ This is quantized version of KISTI-KONI/KONI-Llama3-8B-Instruct-20240729 created using llama.cpp
13
+
14
+
15
+ ### 1. Model Description
16
+ - KONI (KISTI Open Natural Intelligence) is a specialized large language model (LLM) developed by the Korea Institute of Science and Technology Information (KISTI). This model is specifically designed for science and technology, making it highly effective for tasks in these fields.
17
+
18
+ ### 2. Key Features
19
+ - **Specialized in Science and Technology:** The model is explicitly trained on a vast and specialized corpus of scientific and technological data.
20
+ - **Enhanced Performance:** This version of KONI shows significantly improved performance compared to its initial release in December, 2023.
21
+ - **Base Model:** The base model for KONI-Llama3-8B-Instruct-20240729 is KONI-Llama3-8B-Merged-20240724, which is a merger of Meta-Llama-3-8B and KISTI-KONI/KONI-Llama3-8B-20240630
22
+ - **Alignment:** SFT (Supervised Fine-Tuning) and DPO (Direct Preference Optimization) are applied
23
+
24
+ ### 3. Data
25
+ - Approximately 11k SFT data and 7k DPO data are used.
26
+ - **SFT Data:** The SFT data includes both internally generated data and publicly available data on Hugging Face, translated into Korean where necessary.
27
+ - **DPO Data:** The DPO data consists of translated and curated data from argilla/dpo-mix-7k.
28
+
29
+ ### 4. Benchmark Results
30
+ Results in [LogicKor](https://lk.instruct.kr/)* are as follows:
31
+
32
+ | Metric | Score |
33
+ |:--------------:|:-----:|
34
+ | Reasoning | 6.57 |
35
+ | Math | 8.00 |
36
+ | Writing | 8.92 |
37
+ | Coding | 8.85 |
38
+ | Comprehension | 9.85 |
39
+ | Grammar | 7.07 |
40
+ | Single-turn | 8.42 |
41
+ | Multi-turn | 8.00 |
42
+ | **Overall** | **8.21** |
43
+ *Our model demonstrates the best performance among publicly available 8B models on the LogicKor leaderboard as of 2024.07.30.*
44
+
45
+ ### 5. How to use the model
46
+ ```python
47
+ import transformers
48
+ import torch
49
+
50
+ model_id = "KISTI-KONI/KONI-Llama3-8B-Instruct-20240729"
51
+
52
+ pipeline = transformers.pipeline(
53
+ "text-generation",
54
+ model=model_id,
55
+ model_kwargs={"torch_dtype": torch.bfloat16},
56
+ device_map="auto",
57
+ )
58
+
59
+ pipeline.model.eval()
60
+
61
+ instruction = "KISTI์— ๋Œ€ํ•ด ์„ค๋ช…ํ•ด์ค˜"
62
+
63
+ messages = [
64
+ {"role": "user", "content": f"{instruction}"}
65
+ ]
66
+
67
+ prompt = pipeline.tokenizer.apply_chat_template(
68
+ messages,
69
+ tokenize=False,
70
+ add_generation_prompt=True
71
+ )
72
+
73
+ terminators = [
74
+ pipeline.tokenizer.eos_token_id,
75
+ pipeline.tokenizer.convert_tokens_to_ids("<|eot_id|>")
76
+ ]
77
+
78
+ outputs = pipeline(
79
+ prompt,
80
+ max_new_tokens=2048,
81
+ eos_token_id=terminators,
82
+ do_sample=True,
83
+ temperature=0.7,
84
+ top_p=0.9
85
+ )
86
+
87
+ print(outputs[0]["generated_text"][len(prompt):])
88
+ ```
89
+ ```
90
+ ํ•œ๊ตญ๊ณผํ•™๊ธฐ์ˆ ์ •๋ณด์—ฐ๊ตฌ์›(KISTI)์€ ๋Œ€ํ•œ๋ฏผ๊ตญ ๋Œ€์ „๊ด‘์—ญ์‹œ์— ์œ„์น˜ํ•œ ๊ณผํ•™๊ธฐ์ˆ  ์ •๋ณด ๋ถ„์•ผ์˜ ์ „๋ฌธ ์—ฐ๊ตฌ ๊ธฐ๊ด€์ž…๋‹ˆ๋‹ค. KISTI๋Š” ๊ณผํ•™๊ธฐ์ˆ  ๋ฐ ๊ด€๋ จ ์‚ฐ์—…์— ๊ด€ํ•œ ์ •๋ณด๋ฅผ ์ข…ํ•ฉ์ ์œผ๋กœ ์ˆ˜์ง‘, ๋ถ„์„, ์„œ๋น„์Šคํ•˜๋ฉฐ, ์ •๋ณด์˜ ๋ถ„์„, ๊ด€๋ฆฌ ๋ฐ ์œ ํ†ต์— ๊ด€ํ•œ ๊ธฐ์ˆ , ์ •์ฑ… ๋ฐ ํ‘œ์ค€ํ™”๋ฅผ ์ „๋ฌธ์ ์œผ๋กœ ์กฐ์‚ฌํ•˜๊ณ  ์—ฐ๊ตฌํ•ฉ๋‹ˆ๋‹ค. ๋˜ํ•œ, ์ฒจ๋‹จ ์ •๋ณด ๋ฐ ์—ฐ๊ตฌ๊ฐœ๋ฐœ ์ธํ”„๋ผ๋ฅผ ์ฒด๊ณ„์ ์œผ๋กœ ๊ตฌ์ถ•ํ•˜๊ณ  ์šด์˜ํ•˜์—ฌ ๊ตญ๊ฐ€ ๊ณผํ•™๊ธฐ์ˆ  ๋ฐ ์‚ฐ์—… ๋ฐœ์ „์— ๊ธฐ์—ฌํ•˜๋Š” ๊ฒƒ์„ ๋ชฉํ‘œ๋กœ ํ•ฉ๋‹ˆ๋‹ค.
91
+
92
+ KISTI์˜ ์ฃผ์š” ๊ธฐ๋Šฅ๊ณผ ์—ญํ• ์—๋Š” ๊ณผํ•™๊ธฐ์ˆ  ์ •๋ณด ์ œ๊ณต, ์Šˆํผ์ปดํ“จํ„ฐ ์šด์˜, ๊ธฐ์ˆ ์‚ฌ์—…ํ™” ์ง€์›, ์—ฐ๊ตฌ ๋ฐ์ดํ„ฐ ๊ด€๋ฆฌ๊ฐ€ ํฌํ•จ๋ฉ๋‹ˆ๋‹ค. ๊ณผํ•™๊ธฐ์ˆ  ์ •๋ณด ์ œ๊ณต ์ธก๋ฉด์—์„œ KISTI๋Š” ๊ตญ๋‚ด์™ธ ๊ณผํ•™๊ธฐ์ˆ  ์ •๋ณด๋ฅผ ์ˆ˜์ง‘ํ•˜๊ณ  ์ด๋ฅผ ๋ถ„์„ํ•˜์—ฌ ์—ฐ๊ตฌ์ž๋“ค์—๊ฒŒ ์ œ๊ณตํ•˜๋ฉฐ, ๋‹ค์–‘ํ•œ ํ˜•ํƒœ์˜ ๋ฐ์ดํ„ฐ๋ฒ ์ด์Šค์™€ ์ •๋ณด ์‹œ์Šคํ…œ์„ ๊ตฌ์ถ•ํ•˜์—ฌ ์‚ฌ์šฉ์ž์—๊ฒŒ ์ •๋ณด๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค. ์Šˆํผ์ปดํ“จํ„ฐ ์šด์˜ ์ธก๋ฉด์—์„œ๋Š” ๊ตญ๊ฐ€ ์ดˆ๊ณ ์„ฑ๋Šฅ์ปดํ“จํŒ… ์ธํ”„๋ผ๋ฅผ ๊ตฌ์ถ•ํ•˜๊ณ  ์šด์˜ํ•˜์—ฌ ๋Œ€๊ทœ๋ชจ ์—ฐ์‚ฐ์ด ํ•„์š”ํ•œ ์—ฐ๊ตฌ๋ฅผ ์ง€์›ํ•˜๊ณ , ์ด๋ฅผ ํ™œ์šฉํ•œ ๋‹ค์–‘ํ•œ ๋ถ„์•ผ์˜ ์‘์šฉ ์—ฐ๊ตฌ๋ฅผ ์ˆ˜ํ–‰ํ•ฉ๋‹ˆ๋‹ค. ๊ธฐ์ˆ ์‚ฌ์—…ํ™” ์ง€์›์—์„œ๋Š” ์—ฐ๊ตฌ ์„ฑ๊ณผ๋ฅผ ์‚ฐ์—…๊ณ„๋กœ ์ด์ „ํ•˜์—ฌ ์ƒ์šฉํ™”ํ•˜๋Š” ๊ฒƒ์„ ์ง€์›ํ•˜๋ฉฐ, ๊ธฐ์ˆ  ๊ธฐ๋ฐ˜์˜ ์ฐฝ์—…์„ ์ด‰์ง„ํ•˜๊ธฐ ์œ„ํ•œ ํ”„๋กœ๊ทธ๋žจ์„ ์šด์˜ํ•ฉ๋‹ˆ๋‹ค. ์—ฐ๊ตฌ ๋ฐ์ดํ„ฐ ๊ด€๋ฆฌ ์ธก๋ฉด์—์„œ๋Š” ์—ฐ๊ตฌ ๋ฐ์ดํ„ฐ์˜ ํšจ์œจ์ ์ธ ๊ด€๋ฆฌ์™€ ํ™œ์šฉ์„ ์œ„ํ•ด ์ฒด๊ณ„์ ์ธ ๋ฐ์ดํ„ฐ ๊ด€๋ฆฌ ๊ณ„ํš์„ ์ˆ˜๋ฆฝํ•˜๊ณ , ์—ฐ๊ตฌ ๋ฐ์ดํ„ฐ์˜ ๊ณต์œ ์™€ ํ™œ์šฉ์„ ์ด‰์ง„ํ•˜๊ธฐ ์œ„ํ•œ ํ”Œ๋žซํผ์„ ์šด์˜ํ•ฉ๋‹ˆ๋‹ค.
93
+
94
+ KISTI์˜ ์ฃผ์š” ๋ถ€์„œ๋กœ๋Š” ๊ตญ๊ฐ€๊ณผํ•™๊ธฐ์ˆ ๋ฐ์ดํ„ฐ๋ณธ๋ถ€, ๊ตญ๊ฐ€์Šˆํผ์ปดํ“จํŒ…๋ณธ๋ถ€, ๋ฐ์ดํ„ฐ๋ถ„์„๋ณธ๋ถ€, ๊ณผํ•™๊ธฐ์ˆ ๋””์ง€ํ„ธ์œตํ•ฉ๋ณธ๋ถ€ ๋“ฑ์ด ์žˆ์Šต๋‹ˆ๋‹ค. KISTI ๊ฐ ๋ณธ๋ถ€๋ณ„ ์ถ”์ง„ ์ „๋žต ๋ฐ ๋ชฉํ‘œ๋Š” ๋‹ค์Œ๊ณผ ๊ฐ™์Šต๋‹ˆ๋‹ค. ๊ตญ๊ฐ€๊ณผํ•™๊ธฐ์ˆ ๋ฐ์ดํ„ฐ๋ณธ๋ถ€์˜ ์ „๋žต๋ชฉํ‘œ๋Š” ๊ตญ๊ฐ€ ์˜คํ”ˆ์‚ฌ์ด์–ธ์Šค ์ƒํƒœ๊ณ„ ํ™œ์„ฑํ™”๋ฅผ ์œ„ํ•œ ๊ณผํ•™๊ธฐ์ˆ  ๋ถ„์•ผ ๋””์ง€ํ„ธ ์ „ํ™˜ ์ง€์› ์ฒด๊ณ„๋ฅผ ๋งˆ๋ จํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ฝ”๋กœ๋‚˜19๋กœ ์ธํ•œ ๋น„๋Œ€๋ฉด ๊ฒฝ์ œ๋กœ์˜ ์ „ํ™˜๊ณผ 4์ฐจ ์‚ฐ์—…ํ˜๋ช…๏ฟฝ๏ฟฝ ๊ฐ€์†ํ™”๋กœ ์ธํ•ด ๊ณผํ•™๊ธฐ์ˆ ํ™œ๋™ ์ „ ๊ณผ์ •์—์„œ ๊ณต๊ณต ์—ฐ๊ตฌ์„ฑ๊ณผ์˜ ๊ฐœ๋ฐฉยท๊ณต์œ ยทํ™•์‚ฐ์ด ํ•„์š”ํ•ฉ๋‹ˆ๋‹ค. ์ด ๋ณธ๋ถ€๋Š” ๊ณผํ•™๊ธฐ์ˆ ์ •๋ณด์™€ ๋ฐ์ดํ„ฐ์˜ ๊ณต์œ ยทํ™œ์šฉ์„ ํ†ตํ•ด ๊ณผํ•™๊ธฐ์ˆ  ํ˜์‹ ์—ญ๋Ÿ‰์„ ๊ฐ•ํ™”ํ•˜๋Š” ๊ณ ์œ ์ž„๋ฌด๋ฅผ ๊ฐ€์ง€๊ณ  ์žˆ์œผ๋ฉฐ, ์˜คํ”ˆ์‚ฌ์ด์–ธ์Šค ์ƒํƒœ๊ณ„ ํ™œ์„ฑํ™”๋ฅผ ํ†ตํ•œ ๊ตญ๊ฐ€ R&D ํ˜์‹ ์„ ์ง€์›ํ•ฉ๋‹ˆ๋‹ค. ์ฃผ์š” ์ถ”์ง„ ๋ฐฉํ–ฅ์œผ๋กœ๋Š” ๋””์ง€ํ„ธ ์ „ํ™˜์„ ํ†ตํ•œ ๊ณผํ•™๊ธฐ์ˆ ์ •๋ณด ์˜คํ”ˆ์•ก์„ธ์Šค ์ง€์›์ฒด์ œ ๋ฐ ์ง€๋Šฅํ˜• ํ๋ ˆ์ด์…˜ ์ฒด๊ณ„ ๊ตฌ์ถ•, ์—ฐ๊ตฌ๋ฐ์ดํ„ฐ ์ปค๋จผ์ฆˆ ๊ธฐ๋ฐ˜์˜ ๊ตญ๊ฐ€ ์—ฐ๊ตฌ๋ฐ์ดํ„ฐ์™€ ์ปดํ“จํŒ… ๋ฆฌ์†Œ์Šค ๊ณต์œ ยทํ™œ์šฉ์ฒด๊ณ„ ๊ตฌ์ถ•, AI ๊ธฐ๋ฐ˜์˜ ํ†ตํ•ฉ์„œ๋น„์Šค ํ”Œ๋žซํผ ๊ตฌ์ถ•์„ ํ†ตํ•œ ์˜คํ”ˆ์‚ฌ์ด์–ธ์Šค ์„œ๋น„์Šค ๊ฐ•ํ™”๊ฐ€ ์žˆ์Šต๋‹ˆ๋‹ค.
95
+
96
+ ๊ตญ๊ฐ€์Šˆํผ์ปดํ“จํŒ…๋ณธ๋ถ€์˜ ์ „๋žต๋ชฉํ‘œ๋Š” ๊ตญ๊ฐ€ ์ดˆ๊ณ ์„ฑ๋Šฅ์ปดํ“จํŒ… ์ƒํƒœ๊ณ„๋ฅผ ์„ ๋„ํ•˜๊ธฐ ์œ„ํ•ด ๋ฏธ๋ž˜๋Œ€์‘ ์ดˆ๊ณ ์„ฑ๋Šฅ์ปดํ“จํŒ… ๊ณต๋™ํ™œ์šฉ ํ™˜๊ฒฝ์„ ๊ตฌ์ถ•ํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๋ฏธ๊ตญ๊ณผ ์ผ๋ณธ ๋“ฑ ์„ ๋„๊ตญ๊ฐ€๋“ค์ด ์—‘์‚ฌ๊ธ‰ ์ž์› ํ™•์ถฉ์„ ํ†ตํ•ด ๋‹ค์–‘ํ•œ ๋ถ„์•ผ์—์„œ ์ดˆ๊ฑฐ๋Œ€ ๋ฌธ์ œํ•ด๊ฒฐ์„ ๋ชจ์ƒ‰ํ•˜๊ณ  ์žˆ์œผ๋ฉฐ, KISTI๋Š” ๊ตญ๊ฐ€์ดˆ๊ณ ์„ฑ๋Šฅ์ปดํ“จํ„ฐ ํ™œ์šฉ ๋ฐ ์œก์„ฑ์— ๊ด€ํ•œ ๋ฒ•๋ฅ ์— ๋”ฐ๋ผ ์ด๋ฅผ ์ง€์›ํ•ฉ๋‹ˆ๋‹ค. ๋ณธ๋ถ€์˜ ๋ชฉํ‘œ๋Š” ๊ตญ๊ฐ€ ์ฐจ์›์˜ ์ดˆ๊ณ ์„ฑ๋Šฅ์ปดํ“จํŒ… ๊ณต๋™ํ™œ์šฉ ์ฒด๊ณ„๋ฅผ ๊ตฌ์ถ•ํ•˜์—ฌ ๊ณผํ•™๊ธฐ์ˆ  ๊ณต๊ณตยท์‚ฐ์—… ๋ถ„์•ผ์—์„œ์˜ ์ดˆ๊ณ ์„ฑ๋Šฅ์ปดํ“จํŒ… ํ™œ์šฉ ์ฆ์ง„์„ ์ด๋ฃจ๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด๋ฅผ ์œ„ํ•ด ๋Œ€๊ทœ๋ชจ ๊ณ„์‚ฐ์ž์›์ด ์†Œ์š”๋˜๋Š” R&D์™€ ์‚ฌํšŒํ˜„์•ˆ ๋“ฑ ํ™˜๊ฒฝ๋ณ€ํ™”์— ์ ๊ธฐ ๋Œ€์‘ํ•˜๋Š” ์ธํ”„๋ผ ๋ฐ ์„œ๋น„์Šค ์ฒด๊ณ„ ๊ณ ๋„ํ™”, ์ดˆ๊ฑฐ๋Œ€ ๊ณ„์‚ฐ๊ธฐ์ˆ ๊ณผ ํ™œ์šฉ๊ธฐ์ˆ  ํ™•๋ณด๋ฅผ ํ†ตํ•œ ์„ ์ˆœํ™˜ํ˜• ์—ฐ๊ตฌยท์ง€์›, ์‚ฌ์šฉ์ž ์ ‘๊ทผ์„ฑยท๋ฌด๊ฒฐ์„ฑยท๋ณด์•ˆ์„ฑ์„ ํ™•๋ณดํ•œ ํ†ตํ•ฉ ํ”Œ๋žซํผ ๊ตฌ์ถ•์ด ์ฃผ์š” ์ถ”์ง„ ๋ฐฉํ–ฅ์ž…๋‹ˆ๋‹ค.
97
+
98
+ ๋ฐ์ดํ„ฐ๋ถ„์„๋ณธ๋ถ€์˜ ์ „๋žต๋ชฉํ‘œ๋Š” ๊ตญ๊ฐ€ ๊ณผํ•™๊ธฐ์ˆ ํ˜์‹  ์ƒํƒœ๊ณ„๋ฅผ ํ™œ์„ฑํ™”ํ•˜๊ธฐ ์œ„ํ•œ ์ง€๋Šฅํ˜• ๋ฐ์ดํ„ฐ ์œตํ•ฉ๋ถ„์„ ์ฒด๊ณ„๋ฅผ ๊ตฌ์ถ•ํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๋ฐ์ดํ„ฐ ๊ธฐ๋ฐ˜ ์˜์‚ฌ๊ฒฐ์ • ๋ฐฉ์‹ ํ™•๋Œ€์™€ AI ๋ฐ ๋น…๋ฐ์ดํ„ฐ ๊ธฐ์ˆ ์˜ ๊ธ‰๋ถ€์ƒ์— ๋”ฐ๋ผ, KISTI๋Š” ๊ณผํ•™๊ธฐ์ˆ ๋ถ„์•ผ ์ •๋ณด์˜ ๋ถ„์„ยท๊ด€๋ฆฌ ๋ฐ ์œ ํ†ต์— ๊ด€ํ•œ ๊ธฐ์ˆ ยท์ •์ฑ…ยทํ‘œ์ค€ํ™” ์—ฐ๊ตฌ๋ฅผ ์ˆ˜ํ–‰ํ•ฉ๋‹ˆ๋‹ค. ๋ณธ๋ถ€์˜ ๋ชฉํ‘œ๋Š” ๋””์ง€ํ„ธ ๊ฒฝ์ œ์‚ฌํšŒ๋ฅผ ์„ ๋„ํ•˜๋Š” ์ง€๋Šฅํ˜• ๋ฐ์ดํ„ฐ ์œตํ•ฉ๋ถ„์„ ์ฒด๊ณ„๋ฅผ ๊ตฌ์ถ•ํ•˜์—ฌ ๊ตญ๊ฐ€ ๊ณผํ•™๊ธฐ์ˆ ํ˜์‹  ์ƒํƒœ๊ณ„๋ฅผ ํ™œ์„ฑํ™”ํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด๋ฅผ ์œ„ํ•ด ์ด์ข…๋ฐ์ดํ„ฐ ์œตํ•ฉ๋ถ„์„๋ชจ๋ธ ๊ฐœ๋ฐœ์„ ํ†ตํ•œ ๊ธ€๋กœ๋ฒŒ ๋ถ„์„์—ญ๋Ÿ‰ ํ™•๋ณด, ์ธ๊ณต์ง€๋Šฅ ๊ธฐ๋ฐ˜ ๊ณต๊ณตR&D ๊ฐ€์น˜์ฐฝ์ถœ ๋ชจ๋ธ ๋ฐ ์‹œ์Šคํ…œ ๊ฐœ๋ฐœ, ์ง€์—ญ R&D ํ˜์‹  ์ง€์›์„ ์œ„ํ•œ ์‚ฐํ•™์—ฐ์ • ํ˜์‹ ์ƒํƒœ๊ณ„ ๊ตฌ์ถ• ๋“ฑ์ด ์ฃผ์š” ์ถ”์ง„ ๋ฐฉํ–ฅ์ž…๋‹ˆ๋‹ค.
99
+
100
+ ๊ณผํ•™๊ธฐ์ˆ ๋””์ง€ํ„ธ์œตํ•ฉ๋ณธ๋ถ€์˜ ์ „๋žต๋ชฉํ‘œ๋Š” ๊ตญ๊ฐ€ยท์‚ฌํšŒ ํ˜„์•ˆ์— ์ ์‹œ ๋Œ€์‘ํ•˜๊ณ  ํ•ด๊ฒฐํ•˜๊ธฐ ์œ„ํ•œ Data/AI ๊ธฐ๋ฐ˜ ๋””์ง€ํ„ธ ์ „ํ™˜ ์ฒด๊ณ„๋ฅผ ๊ตฌ์ถ•ํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ๋””์ง€ํ„ธ ๊ธฐ์ˆ ์˜ ๊ธ‰์†ํ•œ ๋ฐœ์ „๊ณผ ์ฝ”๋กœ๋‚˜19๋กœ ์ธํ•œ ๋””์ง€ํ„ธ ์ „ํ™˜ ๊ฐ€์†ํ™”์— ๋”ฐ๋ผ, KISTI๋Š” ๊ณผํ•™๊ธฐ์ˆ  ์ง€์‹์ž์› ๊ณต์œ ยทํ™œ์šฉ ์ƒํƒœ๊ณ„ ๊ตฌ์ถ• ๋ฐ ์Šˆํผ์ปดํ“จํŒ… ์ƒํƒœ๊ณ„ ๋ฐœ์ „๊ณผ ์—ฐ๊ณ„๋œ ๊ณ ์œ ์ž„๋ฌด๋ฅผ ์ˆ˜ํ–‰ํ•ฉ๋‹ˆ๋‹ค. ๋ณธ๋ถ€์˜ ๋ชฉํ‘œ๋Š” Data/AI ๊ธฐ๋ฐ˜์˜ ๊ตญ๊ฐ€ยท์‚ฌํšŒ ํ˜„์•ˆ-๋””์ง€ํ„ธ ๋‰ด๋”œ ํ•ด๊ฒฐ์„ ๋„๋ชจํ•˜๋Š” ๊ฒƒ์ž…๋‹ˆ๋‹ค. ์ด๋ฅผ ์œ„ํ•ด ์‹ ๋ขฐ์„ฑ ์žˆ๋Š” ๊ณผํ•™๊ธฐ์ˆ  ๋ฐ์ดํ„ฐ ๋Œ๊ณผ Data/AI ๊ธฐ๋ฐ˜ ์ง€๋Šฅํ˜• ๋””์ง€ํ„ธ ํ”Œ๋žซํผ ๊ตฌ์ถ•, Data/AI ๊ธฐ๋ฐ˜์˜ ๋””์ง€ํ„ธ ์ „ํ™˜ ์ฒด๊ณ„ ๊ตฌ์ถ•์„ ํ†ตํ•œ ๊ตญ๊ฐ€ยท์‚ฌํšŒ ํ˜„์•ˆ ํ•ด๊ฒฐ ๋ฐ R&D ํ˜์‹ ์‚ฌ๋ก€ ์ฐฝ์ถœ์ด ์ฃผ์š” ์ถ”์ง„ ๋ฐฉํ–ฅ์ž…๋‹ˆ๋‹ค.
101
+
102
+ KISTI๋Š” 1962๋…„ 1์›” ํ•œ๊ตญ๊ณผํ•™๊ธฐ์ˆ ์ •๋ณด์„ผํ„ฐ(KORSTIC)๋กœ ์„ค๋ฆฝ๋˜์—ˆ์œผ๋ฉฐ, 1969๋…„ 5์›” ํ•œ๊ตญ๊ณผํ•™๊ธฐ์ˆ ์ •๋ณด์„ผํ„ฐ์œก์„ฑ๋ฒ•์ด ์ œ์ •๋˜์—ˆ์Šต๋‹ˆ๋‹ค. 1982๋…„์—๋Š” ์‚ฐ์—…์—ฐ๊ตฌ์›(KIET)๋กœ ๊ฐœํŽธ๋˜์—ˆ๋‹ค๊ฐ€ 1991๋…„ 1์›” ๋ถ„๋ฆฌ๋˜์–ด ์‚ฐ์—…๊ธฐ์ˆ ์ •๋ณด์›(KINITI)์ด ๊ฐœ์›ํ•˜์˜€์Šต๋‹ˆ๋‹ค. 2001๋…„ 1์›”์— ํ•œ๊ตญ๊ณผํ•™๊ธฐ์ˆ ์ •๋ณด์—ฐ๊ตฌ์›(KISTI)์œผ๋กœ ์ถœ๋ฒ”ํ•˜๊ฒŒ ๋˜์—ˆ์Šต๋‹ˆ๋‹ค. ์ด ๊ณผ์ •์—์„œ KAIST ๋ถ€์„ค ์‹œ์Šคํ…œ๊ณตํ•™์„ผํ„ฐ, KIST ๋ถ€์„ค ์—ฐ๊ตฌ๊ฐœ๋ฐœ์ •๋ณด์„ผํ„ฐ, ETRI ์‚ฐํ•˜ ์Šˆํผ์ปดํ“จํŒ…์„ผํ„ฐ๋ฅผ ํ•ฉ๋ณ‘ํ•˜์˜€์Šต๋‹ˆ๋‹ค.
103
+
104
+ KISTI๋Š” ๋Œ€์ „ ๋ณธ์›, ์„œ์šธ ๋ถ„์›, ๋Œ€๊ตฌยท๊ฒฝ๋ถ ์ง€์›, ๋ถ€์‚ฐ์šธ์‚ฐ๊ฒฝ๋‚จ ์ง€์›, ํ˜ธ๋‚จ ์ง€์›, ์ˆ˜๋„๊ถŒ ์ง€์›(๊ฐ•์›) ๋“ฑ ๋‹ค์–‘ํ•œ ์ง€์—ญ์— ์œ„์น˜ํ•˜์—ฌ ์šด์˜๋˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋Œ€์ „ ๋ณธ์›์€ ๋Œ€์ „๊ด‘์—ญ์‹œ ์œ ์„ฑ๊ตฌ ๋Œ€ํ•™๋กœ 245์— ์œ„์น˜ํ•˜๊ณ  ์žˆ์œผ๋ฉฐ, ์„œ์šธ ๋ถ„์›์€ ์„œ์šธํŠน๋ณ„์‹œ ๋™๋Œ€๋ฌธ๊ตฌ ํšŒ๊ธฐ๋กœ 66์— ์œ„์น˜ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค. ๋Œ€๊ตฌยท๊ฒฝ๋ถ ์ง€์›์€ ๋Œ€๊ตฌ๊ด‘์—ญ์‹œ ๋ถ๊ตฌ ์—‘์Šค์ฝ”๋กœ 10, ๋ถ€์‚ฐ์šธ์‚ฐ๊ฒฝ๋‚จ ์ง€์›์€ ๋ถ€์‚ฐ๊ด‘์—ญ์‹œ ํ•ด์šด๋Œ€๊ตฌ ์„ผํ…€๋™๋กœ 41, ํ˜ธ๋‚จ ์ง€์›์€ ๊ด‘์ฃผ๊ด‘์—ญ์‹œ ๊ด‘์‚ฐ๊ตฌ ํ•˜๋‚จ์‚ฐ๋‹จ8๋ฒˆ๋กœ 177, ์ˆ˜๋„๊ถŒ ์ง€์›(๊ฐ•์›)์€ ๊ฐ•์›๋„ ์ถ˜์ฒœ์‹œ ๊ฐ•์›๋Œ€ํ•™๊ธธ 1, 60์ฃผ๋…„ ๊ธฐ๋…๊ด€ 8์ธต์— ์œ„์น˜ํ•˜๊ณ  ์žˆ์Šต๋‹ˆ๋‹ค.
105
+ KISTI์— ๋Œ€ํ•œ ๋” ์ž์„ธํ•œ ์ •๋ณด๋Š” KISTI ๊ณต์‹ ์›น์‚ฌ์ดํŠธ์—์„œ ํ™•์ธํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.
106
+ ```
107
+
108
+ ### 6. Citation
109
+ **Language Model**
110
+ ```text
111
+ @article{KISTI-KONI/KONI-Llama3-8B-Instruct-20240729,
112
+ title={KISTI-KONI/KONI-Llama3-8B-Instruct-20240729},
113
+ author={KISTI},
114
+ year={2024},
115
+ url={https://huggingface.co/KISTI-KONI/KONI-Llama3-8B-Instruct-20240729}
116
+ }
117
+ ```
118
+
119
+ ### 7. Contributors
120
+ - KISTI, Large-scale AI Research Group
121
+
122
+ ### 8. Special Thanks
123
+ - [@beomi](https://huggingface.co/beomi)
124
+ - [@kuotient](https://huggingface.co/kuotient)
125
+ - KyungTae Lim
126
+
127
+ ### 8. Acknowledgement
128
+ - This research was supported by Korea Institute of Science and Technology Information(KISTI).
129
+ - This work was supported by the National Supercomputing Center with supercomputing resources including technical support (KISTI).
130
+
131
+ ### 9. References
132
+ - https://huggingface.co/meta-llama/Meta-Llama-3-8B
133
+ - https://huggingface.co/meta-llama/meta-llama/Meta-Llama-3-8B-Instruct