RichardErkhov's picture
uploaded readme
a7808b2 verified
Quantization made by Richard Erkhov.
[Github](https://github.com/RichardErkhov)
[Discord](https://discord.gg/pvy7H8DZMG)
[Request more models](https://github.com/RichardErkhov/quant_request)
llama-3-Korean-Bllossom-8B - GGUF
- Model creator: https://huggingface.co/MLP-KTLim/
- Original model: https://huggingface.co/MLP-KTLim/llama-3-Korean-Bllossom-8B/
| Name | Quant method | Size |
| ---- | ---- | ---- |
| [llama-3-Korean-Bllossom-8B.Q2_K.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.Q2_K.gguf) | Q2_K | 2.96GB |
| [llama-3-Korean-Bllossom-8B.IQ3_XS.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.IQ3_XS.gguf) | IQ3_XS | 3.28GB |
| [llama-3-Korean-Bllossom-8B.IQ3_S.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.IQ3_S.gguf) | IQ3_S | 3.43GB |
| [llama-3-Korean-Bllossom-8B.Q3_K_S.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.Q3_K_S.gguf) | Q3_K_S | 3.41GB |
| [llama-3-Korean-Bllossom-8B.IQ3_M.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.IQ3_M.gguf) | IQ3_M | 3.52GB |
| [llama-3-Korean-Bllossom-8B.Q3_K.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.Q3_K.gguf) | Q3_K | 3.74GB |
| [llama-3-Korean-Bllossom-8B.Q3_K_M.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.Q3_K_M.gguf) | Q3_K_M | 3.74GB |
| [llama-3-Korean-Bllossom-8B.Q3_K_L.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.Q3_K_L.gguf) | Q3_K_L | 4.03GB |
| [llama-3-Korean-Bllossom-8B.IQ4_XS.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.IQ4_XS.gguf) | IQ4_XS | 4.18GB |
| [llama-3-Korean-Bllossom-8B.Q4_0.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.Q4_0.gguf) | Q4_0 | 4.34GB |
| [llama-3-Korean-Bllossom-8B.IQ4_NL.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.IQ4_NL.gguf) | IQ4_NL | 4.38GB |
| [llama-3-Korean-Bllossom-8B.Q4_K_S.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.Q4_K_S.gguf) | Q4_K_S | 4.37GB |
| [llama-3-Korean-Bllossom-8B.Q4_K.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.Q4_K.gguf) | Q4_K | 4.58GB |
| [llama-3-Korean-Bllossom-8B.Q4_K_M.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.Q4_K_M.gguf) | Q4_K_M | 4.58GB |
| [llama-3-Korean-Bllossom-8B.Q4_1.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.Q4_1.gguf) | Q4_1 | 4.78GB |
| [llama-3-Korean-Bllossom-8B.Q5_0.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.Q5_0.gguf) | Q5_0 | 5.21GB |
| [llama-3-Korean-Bllossom-8B.Q5_K_S.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.Q5_K_S.gguf) | Q5_K_S | 5.21GB |
| [llama-3-Korean-Bllossom-8B.Q5_K.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.Q5_K.gguf) | Q5_K | 5.34GB |
| [llama-3-Korean-Bllossom-8B.Q5_K_M.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.Q5_K_M.gguf) | Q5_K_M | 5.34GB |
| [llama-3-Korean-Bllossom-8B.Q5_1.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.Q5_1.gguf) | Q5_1 | 5.65GB |
| [llama-3-Korean-Bllossom-8B.Q6_K.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.Q6_K.gguf) | Q6_K | 6.14GB |
| [llama-3-Korean-Bllossom-8B.Q8_0.gguf](https://huggingface.co/RichardErkhov/MLP-KTLim_-_llama-3-Korean-Bllossom-8B-gguf/blob/main/llama-3-Korean-Bllossom-8B.Q8_0.gguf) | Q8_0 | 7.95GB |
Original model description:
---
base_model:
- meta-llama/Meta-Llama-3-8B
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>
# Update!
* ~~[2024.08.09] Llama3.1 버전을 κΈ°λ°˜μœΌλ‘œν•œ Bllossom-8B둜 λͺ¨λΈμ„ μ—…λ°μ΄νŠΈ ν–ˆμŠ΅λ‹ˆλ‹€. κΈ°μ‘΄ llama3기반 Bllossom 보닀 평균 5%정도 μ„±λŠ₯ ν–₯상이 μžˆμ—ˆμŠ΅λ‹ˆλ‹€.~~(μˆ˜μ •μ€‘μ— μžˆμŠ΅λ‹ˆλ‹€.)
* [2024.06.18] μ‚¬μ „ν•™μŠ΅λŸ‰μ„ **250GB**κΉŒμ§€ 늘린 Bllossom ELOλͺ¨λΈλ‘œ μ—…λ°μ΄νŠΈ λ˜μ—ˆμŠ΅λ‹ˆλ‹€. λ‹€λ§Œ 단어확μž₯은 ν•˜μ§€ μ•Šμ•˜μŠ΅λ‹ˆλ‹€. κΈ°μ‘΄ 단어확μž₯된 long-context λͺ¨λΈμ„ ν™œμš©ν•˜κ³  μ‹ΆμœΌμ‹ λΆ„μ€ κ°œμΈμ—°λ½μ£Όμ„Έμš”!
* [2024.06.18] Bllossom ELO λͺ¨λΈμ€ 자체 κ°œλ°œν•œ ELOμ‚¬μ „ν•™μŠ΅ 기반으둜 μƒˆλ‘œμš΄ ν•™μŠ΅λœ λͺ¨λΈμž…λ‹ˆλ‹€. [LogicKor](https://github.com/StableFluffy/LogicKor) 벀치마크 κ²°κ³Ό ν˜„μ‘΄ν•˜λŠ” ν•œκ΅­μ–΄ 10Bμ΄ν•˜ λͺ¨λΈμ€‘ SOTA점수λ₯Ό λ°›μ•˜μŠ΅λ‹ˆλ‹€.
LogicKor μ„±λŠ₯ν‘œ :
| Model | Math | Reasoning | Writing | Coding | Understanding | Grammar | Single ALL | Multi ALL | Overall |
|:---------:|:-----:|:------:|:-----:|:-----:|:----:|:-----:|:-----:|:-----:|:----:|
| gpt-3.5-turbo-0125 | 7.14 | 7.71 | 8.28 | 5.85 | 9.71 | 6.28 | 7.50 | 7.95 | 7.72 |
| gemini-1.5-pro-preview-0215 | 8.00 | 7.85 | 8.14 | 7.71 | 8.42 | 7.28 | 7.90 | 6.26 | 7.08 |
| llama-3-Korean-Bllossom-8B | 5.43 | 8.29 | 9.0 | 4.43 | 7.57 | 6.86 | 6.93 | 6.93 | 6.93 |
# Bllossom | [Demo]() | [Homepage](https://www.bllossom.ai/) | [Github](https://github.com/MLP-Lab/Bllossom) |
<!-- [GPU용 Colab μ½”λ“œμ˜ˆμ œ](https://colab.research.google.com/drive/1fBOzUVZ6NRKk_ugeoTbAOokWKqSN47IG?usp=sharing) | -->
<!-- [CPU용 Colab μ–‘μžν™”λͺ¨λΈ μ½”λ“œμ˜ˆμ œ](https://colab.research.google.com/drive/129ZNVg5R2NPghUEFHKF0BRdxsZxinQcJ?usp=drive_link) -->
```bash
저희 BllossomνŒ€ μ—μ„œ ν•œκ΅­μ–΄-μ˜μ–΄ 이쀑 μ–Έμ–΄λͺ¨λΈμΈ Bllossom을 κ³΅κ°œν–ˆμŠ΅λ‹ˆλ‹€!
μ„œμšΈκ³ΌκΈ°λŒ€ μŠˆνΌμ»΄ν“¨νŒ… μ„Όν„°μ˜ μ§€μ›μœΌλ‘œ 100GBκ°€λ„˜λŠ” ν•œκ΅­μ–΄λ‘œ λͺ¨λΈμ „체λ₯Ό ν’€νŠœλ‹ν•œ ν•œκ΅­μ–΄ κ°•ν™” 이쀑언어 λͺ¨λΈμž…λ‹ˆλ‹€!
ν•œκ΅­μ–΄ μž˜ν•˜λŠ” λͺ¨λΈ μ°Ύκ³  μžˆμ§€ μ•ŠμœΌμ…¨λ‚˜μš”?
- ν•œκ΅­μ–΄ 졜초! 무렀 3λ§Œκ°œκ°€ λ„˜λŠ” ν•œκ΅­μ–΄ μ–΄νœ˜ν™•μž₯
- Llama3λŒ€λΉ„ λŒ€λž΅ 25% 더 κΈ΄ 길이의 ν•œκ΅­μ–΄ Context μ²˜λ¦¬κ°€λŠ₯
- ν•œκ΅­μ–΄-μ˜μ–΄ Pararell Corpusλ₯Ό ν™œμš©ν•œ ν•œκ΅­μ–΄-μ˜μ–΄ 지식연결 (μ‚¬μ „ν•™μŠ΅)
- ν•œκ΅­μ–΄ λ¬Έν™”, μ–Έμ–΄λ₯Ό κ³ λ €ν•΄ μ–Έμ–΄ν•™μžκ°€ μ œμž‘ν•œ 데이터λ₯Ό ν™œμš©ν•œ λ―Έμ„Έμ‘°μ •
- κ°•ν™”ν•™μŠ΅
이 λͺ¨λ“ κ²Œ ν•œκΊΌλ²ˆμ— 적용되고 상업적 이용이 κ°€λŠ₯ν•œ Bllossom을 μ΄μš©ν•΄ μ—¬λŸ¬λΆ„ 만의 λͺ¨λΈμ„ λ§Œλ“€μ–΄λ³΄μ„Έμš₯!
무렀 Colab 무료 GPU둜 ν•™μŠ΅μ΄ κ°€λŠ₯ν•©λ‹ˆλ‹€. ν˜Ήμ€ μ–‘μžν™” λͺ¨λΈλ‘œ CPUμ—μ˜¬λ €λ³΄μ„Έμš” [μ–‘μžν™”λͺ¨λΈ](https://huggingface.co/MLP-KTLim/llama-3-Korean-Bllossom-8B-4bit)
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™‚
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”!!)
3. Bllossom은 NAACL2024, LREC-COLING2024 (ꡬ두) λ°œν‘œλ‘œ μ±„νƒλ˜μ—ˆμŠ΅λ‹ˆλ‹€.
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‹€!!
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”.
```
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;">
<!-- 첫 번째 컬럼 -->
<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>
<!-- 두 번째 컬럼 (ν•„μš”ν•˜λ‹€λ©΄) -->
<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)γ…€γ…€γ…€γ…€γ…€γ…€γ…€γ…€</p>
</div>
</div>
# NEWS
* [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.
* [2024.05.08] Vocab Expansion Model Update
* [2024.04.25] We released Bllossom v2.0, based on llama-3
## 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 = "MLP-KTLim/llama-3-Korean-Bllossom-8B"
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. 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”.'''
instruction = "μ„œμšΈμ˜ 유λͺ…ν•œ κ΄€κ΄‘ μ½”μŠ€λ₯Ό λ§Œλ“€μ–΄μ€„λž˜?"
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):])
```
```
# 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”.
### μ½”μŠ€ 1: 역사와 λ¬Έν™” 탐방
1. **경볡ꢁ**
- 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‹€.
2. **뢁촌 ν•œμ˜₯λ§ˆμ„**
- 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‹€.
3. **인사동**
- 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‹€.
4. **μ²­κ³„μ²œ**
- μ„œμšΈμ˜ 쀑심에 μœ„μΉ˜ν•œ 천문으둜, μ‘°κΉ…κ³Ό 산책을 즐길 수 μžˆλŠ” κ³³μž…λ‹ˆλ‹€.
### μ½”μŠ€ 2: μžμ—°κ³Ό μ‡Όν•‘
1. **남산 μ„œμšΈνƒ€μ›Œ**
- 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‹€.
2. **λͺ…동**
- 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‹€.
3. **ν•œκ°•κ³΅μ›**
- μ„œμšΈμ˜ μ£Όμš” 곡원 쀑 ν•˜λ‚˜λ‘œ, μ‘°κΉ…, μžμ „κ±° 타기, λ°°λ‚­ 여행을 즐길 수 μžˆμŠ΅λ‹ˆλ‹€.
4. **ν™λŒ€**
- 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‹€.
### μ½”μŠ€ 3: ν˜„λŒ€μ™€ μ „ν†΅μ˜ μ‘°ν™”
1. **λ™λŒ€λ¬Έ λ””μžμΈ ν”ŒλΌμž (DDP)**
- 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‹€.
2. **μ΄νƒœμ›**
- 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‹€.
3. **κ΄‘ν™”λ¬Έ**
- μ„œμšΈμ˜ 쀑심에 μœ„μΉ˜ν•œ κ΄‘μž₯으둜, λ‹€μ–‘ν•œ 곡연과 행사가 μ—΄λ¦½λ‹ˆλ‹€.
4. **μ„œμšΈλžœλ“œ**
- 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‹€.
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”!
```
### Python code with AutoModel
```python
import os
import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
model_id = 'MLP-KTLim/llama-3-Korean-Bllossom-8B'
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. 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”.'''
instruction = "μ„œμšΈμ˜ 유λͺ…ν•œ κ΄€κ΄‘ μ½”μŠ€λ₯Ό λ§Œλ“€μ–΄μ€„λž˜?"
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))
```
```
# 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”.
### μ½”μŠ€ 1: 역사와 λ¬Έν™” 탐방
1. **경볡ꢁ**
- 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‹€.
2. **뢁촌 ν•œμ˜₯λ§ˆμ„**
- 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‹€.
3. **인사동**
- 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‹€.
4. **μ²­κ³„μ²œ**
- μ„œμšΈμ˜ 쀑심에 μœ„μΉ˜ν•œ 천문으둜, μ‘°κΉ…κ³Ό 산책을 즐길 수 μžˆλŠ” κ³³μž…λ‹ˆλ‹€.
### μ½”μŠ€ 2: μžμ—°κ³Ό μ‡Όν•‘
1. **남산 μ„œμšΈνƒ€μ›Œ**
- 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‹€.
2. **λͺ…동**
- 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‹€.
3. **ν•œκ°•κ³΅μ›**
- μ„œμšΈμ˜ μ£Όμš” 곡원 쀑 ν•˜λ‚˜λ‘œ, μ‘°κΉ…, μžμ „κ±° 타기, λ°°λ‚­ 여행을 즐길 수 μžˆμŠ΅λ‹ˆλ‹€.
4. **ν™λŒ€**
- 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‹€.
### μ½”μŠ€ 3: ν˜„λŒ€μ™€ μ „ν†΅μ˜ μ‘°ν™”
1. **λ™λŒ€λ¬Έ λ””μžμΈ ν”ŒλΌμž (DDP)**
- 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‹€.
2. **μ΄νƒœμ›**
- 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‹€.
3. **κ΄‘ν™”λ¬Έ**
- μ„œμšΈμ˜ 쀑심에 μœ„μΉ˜ν•œ κ΄‘μž₯으둜, λ‹€μ–‘ν•œ 곡연과 행사가 μ—΄λ¦½λ‹ˆλ‹€.
4. **μ„œμšΈλžœλ“œ**
- 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‹€.
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”!
```
## 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
- μž„κ²½νƒœ(KyungTae Lim), Professor at Seoultech. `[email protected]`
- ν•¨μ˜κ· (Younggyun Hahm), CEO of Teddysum. `[email protected]`
- κΉ€ν•œμƒ˜(Hansaem Kim), Professor at Yonsei. `[email protected]`
## Contributor
- 졜창수(Chansu Choi), [email protected]
- 김상민(Sangmin Kim), [email protected]
- μ›μΈν˜Έ(Inho Won), [email protected]
- κΉ€λ―Όμ€€(Minjun Kim), [email protected]
- μ†‘μŠΉμš°(Seungwoo Song), [email protected]
- μ‹ λ™μž¬(Dongjae Shin), [email protected]
- μž„ν˜„μ„(Hyeonseok Lim), [email protected]
- μœ‘μ •ν›ˆ(Jeonghun Yuk), [email protected]
- μœ ν•œκ²°(Hangyeol Yoo), [email protected]
- μ†‘μ„œν˜„(Seohyun Song), [email protected]