Edit model card

Bllossom | Demo | Homepage | Github | Colab-tutorial |

저희 Bllossom ν”„λ‘œμ νŠΈ νŒ€μ—μ„œ ν•œκ΅­μ–΄-μ˜μ–΄ 이쀑 μ–Έμ–΄λͺ¨λΈμΈ Bllossom-70.8Bλ₯Ό 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₯!
GPUκ°€ λΆ€μ‘±ν•˜λ©΄ μ–‘μžν™” λͺ¨λΈλ‘œ λ°”λ‘œ μ„œλΉ„μŠ€λ₯Ό ν™œμš©ν•΄ λ³΄μ„Έμš” [μ–‘μžν™”λͺ¨λΈ](https://huggingface.co/Bllossom/llama-3-Korean-Bllossom-70B-gguf-Q4_K_M)!!

1. Bllossom-70.8BλŠ” 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, Teddysum and Yonsei Univ

Demo Video

Bllossom-V Demo

Bllossom Demo(Kakao)γ…€γ…€γ…€γ…€γ…€γ…€γ…€γ…€

NEWS

  • [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

Install Dependencies

pip install torch transformers==4.40.0 accelerate

Python code with Pipeline

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 = '''당신은 μœ μš©ν•œ AI μ–΄μ‹œμŠ€ν„΄νŠΈμž…λ‹ˆλ‹€. μ‚¬μš©μžμ˜ μ§ˆμ˜μ— λŒ€ν•΄ μΉœμ ˆν•˜κ³  μ •ν™•ν•˜κ²Œ λ‹΅λ³€ν•΄μ•Ό ν•©λ‹ˆλ‹€.
You are a helpful AI assistant, you'll need to answer users' queries in a friendly and accurate manner.'''
instruction = "μ„œμšΈκ³Όν•™κΈ°μˆ λŒ€ν•™κ΅ MLP연ꡬ싀에 λŒ€ν•΄ μ†Œκ°œν•΄μ€˜"

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):])

# μ„œμšΈκ³Όν•™κΈ°μˆ λŒ€ν•™κ΅ MLP연ꡬ싀은 λ©€ν‹°λͺ¨λ‹¬ μžμ—°μ–΄μ²˜λ¦¬ 연ꡬλ₯Ό ν•˜κ³  μžˆμŠ΅λ‹ˆλ‹€. ꡬ성원은 μž„κ²½νƒœ κ΅μˆ˜μ™€ κΉ€λ―Όμ€€, 김상민, 졜창수, μ›μΈν˜Έ, μœ ν•œκ²°, μž„ν˜„μ„, μ†‘μŠΉμš°, μœ‘μ •ν›ˆ, μ‹ λ™μž¬ 학생이 μžˆμŠ΅λ‹ˆλ‹€.

Python code with AutoModel


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 = '''당신은 μœ μš©ν•œ AI μ–΄μ‹œμŠ€ν„΄νŠΈμž…λ‹ˆλ‹€. μ‚¬μš©μžμ˜ μ§ˆμ˜μ— λŒ€ν•΄ μΉœμ ˆν•˜κ³  μ •ν™•ν•˜κ²Œ λ‹΅λ³€ν•΄μ•Ό ν•©λ‹ˆλ‹€.
You are a helpful AI assistant, you'll need to answer users' queries in a friendly and accurate manner.'''
instruction = "μ„œμšΈκ³Όν•™κΈ°μˆ λŒ€ν•™κ΅ MLP연ꡬ싀에 λŒ€ν•΄ μ†Œκ°œν•΄μ€˜"

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))
# μ„œμšΈκ³Όν•™κΈ°μˆ λŒ€ν•™κ΅ MLP연ꡬ싀은 λ©€ν‹°λͺ¨λ‹¬ μžμ—°μ–΄μ²˜λ¦¬ 연ꡬλ₯Ό ν•˜κ³  μžˆμŠ΅λ‹ˆλ‹€. ꡬ성원은 μž„κ²½νƒœ κ΅μˆ˜μ™€ κΉ€λ―Όμ€€, 김상민, 졜창수, μ›μΈν˜Έ, μœ ν•œκ²°, μž„ν˜„μ„, μ†‘μŠΉμš°, μœ‘μ •ν›ˆ, μ‹ λ™μž¬ 학생이 μžˆμŠ΅λ‹ˆλ‹€.

Citation

Language Model

@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

@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. ktlim@seoultech.ac.kr
  • ν•¨μ˜κ· (Younggyun Hahm), CEO of Teddysum. hahmyg@teddysum.ai
  • κΉ€ν•œμƒ˜(Hansaem Kim), Professor at Yonsei. khss@yonsei.ac.kr

Contributor

Downloads last month
1,116
Safetensors
Model size
70.8B params
Tensor type
BF16
Β·
Inference API
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for Bllossom/llama-3-Korean-Bllossom-70B

Finetuned
this model
Finetunes
2 models
Merges
2 models
Quantizations
1 model