Ko-Llama3-Luxia-8B / README.md
sangmine's picture
Update README.md
f3cd592 verified
|
raw
history blame
8.36 kB
metadata
license: llama3
language:
  - en
  - ko
pipeline_tag: text-generation
tags:
  - saltlux
  - luxia
  - meta
  - llama-3
  - pytorch

Model Details

Saltlux, AI Labs ์–ธ์–ด๋ชจ๋ธํŒ€์—์„œ ํ•™์Šต ๋ฐ ๊ณต๊ฐœํ•œ Ko-Llama3-Luxia-8B ๋ชจ๋ธ์€ Meta์—์„œ ์ถœ์‹œํ•œ Llama-3-8B ๋ชจ๋ธ์„ ํ•œ๊ตญ์–ด์— ํŠนํ™”ํ•œ ๋ชจ๋ธ์ž…๋‹ˆ๋‹ค.

์ž์ฒด ๋ณด์œ ํ•˜๊ณ  ์žˆ๋Š” 1TB ์ด์ƒ์˜ ํ•œ๊ตญ์–ด ํ•™์Šต ๋ฐ์ดํ„ฐ ์ค‘, ์•ฝ 100GB ์ •๋„์˜ ๋ฐ์ดํ„ฐ๋ฅผ ์„ ๋ณ„ํ•˜์—ฌ ์‚ฌ์ „ํ•™์Šต์— ํ™œ์šฉํ•˜์˜€์Šต๋‹ˆ๋‹ค.

๋˜ํ•œ ๊ณต๊ฐœ๋œ Llama-3 Tokenizer๋ฅผ ํ•œ๊ตญ์–ด๋กœ ํ™•์žฅํ•˜๊ณ  ์‚ฌ์ „ํ•™์Šต์— ํ™œ์šฉํ–ˆ์Šต๋‹ˆ๋‹ค.

  • Meta Llama-3: Meta developed and released the Meta Llama 3 family of large language models (LLMs), a collection of pretrained and instruction tuned generative text models in 8 and 70B sizes. The Llama 3 instruction tuned models are optimized for dialogue use cases and outperform many of the available open source chat models on common industry benchmarks. Further, in developing these models, we took great care to optimize helpfulness and safety.
  • License: Llama3 License https://llama.meta.com/llama3/license

Intended Use

Ko-Llama3-Luxia-8B๋Š” ์—ฐ๊ตฌ์šฉ์œผ๋กœ ์ œ์ž‘๋˜์—ˆ์œผ๋ฉฐ, ๋‹ค์–‘ํ•œ ์ž์—ฐ์–ด ์ƒ์„ฑ ํƒœ์Šคํฌ๋ฅผ ์œ„ํ•ด ์ž์œ ๋กญ๊ฒŒ ํ•™์Šต ๋ฐ ํ™œ์šฉํ•  ์ˆ˜ ์žˆ์Šต๋‹ˆ๋‹ค.

How to Use

ํ•ด๋‹น ๋ชจ๋ธ ์นด๋“œ์—๋Š” Ko-Llama3-Luxia-8B ๋ชจ๋ธ๊ณผ transformers ๋ผ์ด๋ธŒ๋Ÿฌ๋ฆฌ ๊ธฐ๋ฐ˜์˜ ์˜ˆ์‹œ ์ฝ”๋“œ๋ฅผ ์ œ๊ณตํ•ฉ๋‹ˆ๋‹ค.

import transformers
import torch

model_id = "saltlux/Ko-Llama3-Luxia-8B"

pipeline = transformers.pipeline(
    "text-generation", model=model_id, model_kwargs={"torch_dtype": torch.bfloat16}, device_map="auto"
)
pipeline("<|begin_of_text|>์•ˆ๋…•ํ•˜์„ธ์š”. ์†”ํŠธ๋ฃฉ์Šค AI Labs ์ž…๋‹ˆ๋‹ค.")

Training Details

ํ•œ๊ตญ์–ด ํŠนํ™”๋ฅผ ์œ„ํ•œ ์‚ฌ์ „ํ•™์Šต ๋ฐ์ดํ„ฐ๋Š” Saltlux์—์„œ ๋ณด์œ ํ•œ ๋‰ด์Šค, ๋ฒ•๋ฅ , ํŠนํ—ˆ, ์˜๋ฃŒ, ์—ญ์‚ฌ, ์‚ฌํšŒ, ๋ฌธํ™”, ๋Œ€ํ™”(๋ฌธ์–ด/๊ตฌ์–ด) ๋“ฑ์˜ ๋„๋ฉ”์ธ์œผ๋กœ ๊ตฌ์„ฑ๋œ 100GB ์ˆ˜์ค€์˜ ์ฝ”ํผ์Šค(~2023๋…„)๋ฅผ ํ™œ์šฉํ•˜์˜€์Šต๋‹ˆ๋‹ค.

Use Device

์‚ฌ์ „ํ•™์Šต์€ NVIDIA H100 80GB * 8EA ์žฅ๋น„๋ฅผ ํ™œ์šฉํ•˜์—ฌ ์ง„ํ–‰ํ•˜์˜€์Šต๋‹ˆ๋‹ค.

Training Hyperparameters

Model Params Context length GQA Learning rate Batch Precision
Ko-Llama3-Luxia-8B 8B 8k yes 1e-5 128 bf16

Tokenizer

Llama-3-Tokenizer๋ฅผ ํ•œ๊ตญ์–ด ํŠนํ™”ํ•˜๊ธฐ ์œ„ํ•ด ํ•œ๊ตญ์–ด ํ† ํฐ 17,536๊ฐœ๋ฅผ ์ถ”๊ฐ€ํ•˜๊ณ  ํ™œ์šฉํ•˜์˜€์Šต๋‹ˆ๋‹ค.

Model Vocab Size
Llama-3 128,256
Ko-Llama3-Luxia-8B 145,792

Tokenizer Result

  • Ko

    ์ž…๋ ฅ Llama-3 Ko-Llama3-Luxia-8B
    ์š”์ฆ˜ ๋‚ ์”จ๊ฐ€ ๋„ˆ๋ฌด ์˜ค๋ฝ๊ฐ€๋ฝํ•ด์„œ ์•„์ง๋„ ๊ฒจ์šธ์˜ท์„ ๋ชป์น˜์› ์–ด์š”.. ['์š”', '์ฆ˜', ' ๋‚ ', '์”จ', '๊ฐ€', ' ๋„ˆ๋ฌด', ' ์˜ค', '๋ฝ', '๊ฐ€', '๋ฝ', 'ํ•ด์„œ', ' ์•„์ง', '๋„', ' ๊ฒจ', '์šธ', '๏ฟฝ', '๏ฟฝ', '์„', ' ๋ชป', '์น˜', '์› ', '์–ด์š”', '..'] ['์š”์ฆ˜', ' ๋‚ ์”จ', '๊ฐ€', ' ๋„ˆ๋ฌด', ' ์˜ค๋ฝ', '๊ฐ€๋ฝ', 'ํ•ด์„œ', ' ์•„์ง', '๋„', ' ๊ฒจ์šธ', '์˜ท', '์„', ' ๋ชป', '์น˜', '์› ', '์–ด์š”', '..']
    ๋ง›์žˆ๋Š” ๋ฐฅ์„ ๋“œ์…จ์Šต๋‹ˆ๊นŒ? ๋ง›์ด ๊ถ๊ธˆํ•˜๋„ค์š”. ['๋ง›', '์žˆ๋Š”', ' ๏ฟฝ', '๏ฟฝ', '์„', ' ๋“œ', '์…จ', '์Šต', '๋‹ˆ๊นŒ', '?', ' ๋ง›', '์ด', ' ๊ถ๊ธˆ', 'ํ•˜', '๋„ค์š”', '.'] ['๋ง›', '์žˆ๋Š”', ' ๋ฐฅ', '์„', ' ๋“œ์…จ', '์Šต', '๋‹ˆ๊นŒ', '?', ' ๋ง›', '์ด', ' ๊ถ๊ธˆ', 'ํ•˜', '๋„ค์š”', '.']
    ๋Œ€๋ฒ•์›๋ถ€ํ„ฐ ํ•˜๊ธ‰์‹ฌ ํŒ๋ก€๊นŒ์ง€ ์›ํ•˜๋Š” ํŒ๋ก€๋ฅผ ์ฐพ๋Š” ๊ฐ€์žฅ ๋น ๋ฅธ ๋ฐฉ๋ฒ• - ์„œ๋ฉด ๊ฒ€์ƒ‰, ์š”์ฒญ ํŒ๋ก€, ์œ ์‚ฌ ํŒ๋ก€, AI ์ถ”์ฒœ, ํŒ๋ก€ ๋ฐ ๋ฒ•๋ น ๊ฒ€์ƒ‰. ['๋Œ€', '๋ฒ•', '์›', '๋ถ€ํ„ฐ', ' ํ•˜', '๊ธ‰', '์‹ฌ', ' ํŒ', '๋ก€', '๊นŒ์ง€', ' ์›', 'ํ•˜๋Š”', ' ํŒ', '๋ก€', '๋ฅผ', ' ์ฐพ', '๋Š”', ' ๊ฐ€์žฅ', ' ๋น ', '๋ฅธ', ' ๋ฐฉ๋ฒ•', ' -', ' ์„œ', '๋ฉด', ' ๊ฒ€์ƒ‰', ',', ' ์š”์ฒญ', ' ํŒ', '๋ก€', ',', ' ์œ ', '์‚ฌ', ' ํŒ', '๋ก€', ',', ' AI', ' ์ถ”์ฒœ', ',', ' ํŒ', '๋ก€', ' ๋ฐ', ' ๋ฒ•', '๋ น', ' ๊ฒ€์ƒ‰', '.'] ['๋Œ€', '๋ฒ•', '์›', '๋ถ€ํ„ฐ', ' ํ•˜', '๊ธ‰', '์‹ฌ', ' ํŒ๋ก€', '๊นŒ์ง€', ' ์›', 'ํ•˜๋Š”', ' ํŒ๋ก€', '๋ฅผ', ' ์ฐพ', '๋Š”', ' ๊ฐ€์žฅ', ' ๋น ๋ฅธ', ' ๋ฐฉ๋ฒ•', ' -', ' ์„œ๋ฉด', ' ๊ฒ€์ƒ‰', ',', ' ์š”์ฒญ', ' ํŒ๋ก€', ',', ' ์œ ์‚ฌ', ' ํŒ๋ก€', ',', ' AI', ' ์ถ”์ฒœ', ',', ' ํŒ๋ก€', ' ๋ฐ', ' ๋ฒ•๋ น', ' ๊ฒ€์ƒ‰', '.']
    ๋ณธ ๋ฐœ๋ช…์€ ๊ธˆ์†ํŒ์˜ ๋‹ค์ˆ˜ ๋ถ€๋ถ„์„ ์—์นญ์‹œ์ผœ ํŠน์ • ๋ฌด๋Šฌ๋ชจ์–‘์„ ํ˜•์„ฑํ•˜๋Š” ๊ฑด์ถ•์šฉ ๊ธˆ์†์žฌ ์žฅ์‹ํŒ์œผ๋กœ ์ด๋ฃจ์–ด์ง„ ๊ฒƒ์— ํŠน์ง•์ด ์žˆ๋‹ค. ['๋ณธ', ' ๋ฐœ', '๋ช…', '์€', ' ๊ธˆ', '์†', 'ํŒ', '์˜', ' ๋‹ค', '์ˆ˜', ' ๋ถ€๋ถ„', '์„', ' ์—', '์นญ', '์‹œ', '์ผœ', ' ํŠน', '์ •', ' ๋ฌด', '๏ฟฝ', '๏ฟฝ', '๋ชจ', '์–‘', '์„', ' ํ˜•', '์„ฑ', 'ํ•˜๋Š”', ' ๊ฑด', '์ถ•', '์šฉ', ' ๊ธˆ', '์†', '์žฌ', ' ์žฅ', '์‹', 'ํŒ', '์œผ๋กœ', ' ์ด๋ฃจ', '์–ด์ง„', ' ๊ฒƒ', '์—', ' ํŠน', '์ง•', '์ด', ' ์žˆ๋‹ค', '.'] ['๋ณธ', ' ๋ฐœ๋ช…', '์€', ' ๊ธˆ์†', 'ํŒ', '์˜', ' ๋‹ค์ˆ˜', ' ๋ถ€๋ถ„', '์„', ' ์—์นญ', '์‹œ', '์ผœ', ' ํŠน์ •', ' ๋ฌด๋Šฌ', '๋ชจ', '์–‘', '์„', ' ํ˜•์„ฑ', 'ํ•˜๋Š”', ' ๊ฑด์ถ•', '์šฉ', ' ๊ธˆ์†', '์žฌ', ' ์žฅ์‹', 'ํŒ', '์œผ๋กœ', ' ์ด๋ฃจ์–ด์ง„', ' ๊ฒƒ', '์—', ' ํŠน์ง•', '์ด', ' ์žˆ๋‹ค', '.']
    ๊ณจ๋‹ค๊ณต์ฆ์€ ์™œ ์ƒ๊ธฐ๋Š”๊ฑฐ์—์š”? ๊ทธ๋ฆฌ๊ณ  ์น˜๋ฃŒํ•˜๋ ค๋ฉด ์–ด๋–ป๊ฒŒํ•ด์•ผํ•˜์ฃ ? ['๊ณจ', '๋‹ค', '๊ณต', '์ฆ', '์€', ' ์™œ', ' ์ƒ', '๊ธฐ๋Š”', '๊ฑฐ', '์—', '์š”', '?', ' ๊ทธ๋ฆฌ๊ณ ', ' ์น˜', '๋ฃŒ', 'ํ•˜๋ ค', '๋ฉด', ' ์–ด๋–ป๊ฒŒ', 'ํ•ด์•ผ', 'ํ•˜', '์ฃ ', '?'] ['๊ณจ', '๋‹ค', '๊ณต์ฆ', '์€', ' ์™œ', ' ์ƒ', '๊ธฐ๋Š”', '๊ฑฐ', '์—', '์š”', '?', ' ๊ทธ๋ฆฌ๊ณ ', ' ์น˜๋ฃŒ', 'ํ•˜๋ ค', '๋ฉด', ' ์–ด๋–ป๊ฒŒ', 'ํ•ด์•ผ', 'ํ•˜', '์ฃ ', '?']
  • En

    ์ž…๋ ฅ Llama-3 Ko-Llama3-Luxia-8B
    Korean cuisine, hanguk yori, or hansik, has evolved through centuries of social and political change. ['K', 'orean', ' cuisine', ',', ' h', 'angu', 'k', ' y', 'ori', ',', ' or', ' hans', 'ik', ',', ' has', ' evolved', ' through', ' centuries', ' of', ' social', ' and', ' political', ' change', '.'] ['K', 'orean', ' cuisine', ',', ' h', 'angu', 'k', ' y', 'ori', ',', ' or', ' hans', 'ik', ',', ' has', ' evolved', ' through', ' centuries', ' of', ' social', ' and', ' political', ' change', '.']
    Son Heung-min is a South Korean professional footballer who plays as a forward for and captains both Premier League club Tottenham Hotspur and the South Korea national team. ['Son', ' He', 'ung', '-min', ' is', ' a', ' South', ' Korean', ' professional', ' football', 'er', ' who', ' plays', ' as', ' a', ' forward', ' for', ' and', ' captains', ' both', ' Premier', ' League', ' club', ' Tottenham', ' Hot', 'sp', 'ur', ' and', ' the', ' South', ' Korea', ' national', ' team', '.'] ['Son', ' He', 'ung', '-min', ' is', ' a', ' South', ' Korean', ' professional', ' football', 'er', ' who', ' plays', ' as', ' a', ' forward', ' for', ' and', ' captains', ' both', ' Premier', ' League', ' club', ' Tottenham', ' Hot', 'sp', 'ur', ' and', ' the', ' South', ' Korea', ' national', ' team', '.']

Citation instructions

Ko-Llama3-Luxia-8B

@article{kollama3luxiamodelcard,
  title={Ko Llama 3 Luxia Model Card},
  author={AILabs@Saltux},
  year={2024},
  url={https://huggingface.co/saltlux/Ko-Llama3-Luxia-8B/blob/main/README.md}
}

Original Llama-3

@article{llama3modelcard,
title={Llama 3 Model Card},
author={AI@Meta},
year={2024},
url={https://github.com/meta-llama/llama3/blob/main/MODEL_CARD.md}
}