Gemma-2-2b μ΄ˆλ“±ν•™μƒ κΈ€ λ³€ν™˜κΈ°

πŸ’» Model Description

  • Gemma2-2b ν•œκ΅­ μ΄ˆλ“±ν•™μƒ κΈ€ λ³€ν™˜κΈ°λŠ” μž…λ ₯ν•œ 글을 μ΄ˆλ“±ν•™μƒμ΄ μ“΄ 것 같은 κΈ€λ‘œ λ³€ν™˜ν•΄ μ£ΌλŠ” λͺ¨λΈμž…λ‹ˆλ‹€.
  • Gemma2-2b-it λͺ¨λΈμ„ base model둜 μ‚¬μš©ν•˜μ˜€κ³  LoRA기법을 μ‚¬μš©ν•˜μ—¬ 효율적으둜 fine-tuning ν•˜μ˜€μŠ΅λ‹ˆλ‹€.

(Gemma2-2b is a model that transforms input text to resemble the writing style of an elementary school student. It is based on the Gemma2-2b-it model and was fine-tuned efficiently using the LoRA technique.)

🚦Usage

import transformers
from huggingface_hub import notebook_login

notebook_login()

BASE_MODEL = "skwh54/Gemma-2-2b-it-elementary-style-document"

model = AutoModelForCausalLM.from_pretrained(BASE_MODEL, device_map={"":0})
tokenizer = AutoTokenizer.from_pretrained(BASE_MODEL)

πŸ“‚ Example

Input Output
μ•ˆλ…•ν•˜μ„Έμš”. μ €λŠ” μΉœκ΅¬μ™€ κ³΅μ›μ—μ„œ λ§Œλ‚˜μ„œ λ†€μ•˜μ–΄μš”. 날씨가 μ’‹μ•„μ„œ 놀기 μ’‹μ•˜μ–΄μš”. λ§›μžˆλŠ” μŒμ‹λ“€λ„ 많이 λ¨Ήμ—ˆμ–΄μš”. 내일도 놀고 μ‹Άλ‹€! γ…Žγ…‡ λ‚œ μΉœκ΅¬λž‘ κ³΅μ›μ—μ„œ λ§Œλ‚˜μ„œ λ†€μ•˜λŠ”λ° 날씨가 μ’‹μ•„μ„œ 더 μ’‹μ•˜κ³  λ§›μžˆλŠ” μŒμ‹λ„ 많이 λ¨Ήμ—ˆμŒ 내일도 놀고 μ‹Άλ‹€
μ˜€λŠ˜μ€ νšŒμ‚¬μ—μ„œ μ€‘μš”ν•œ ν”„λ‘œμ νŠΈ 회의λ₯Ό ν–ˆλ‹€. μ€€λΉ„ν•œ 만큼 쒋은 ν”Όλ“œλ°±μ„ λ°›μ•„μ„œ λΏŒλ“―ν–ˆλ‹€. 퇴근 ν›„μ—λŠ” μš΄λ™μ„ ν•˜κ³  λ‚˜μ„œ μ§‘μ—μ„œ 책을 읽으며 ν•˜λ£¨λ₯Ό λ§ˆλ¬΄λ¦¬ν–ˆλ‹€. 였늘 νšŒμ‚¬μ—μ„œ μ€‘μš”ν•œ ν”„λ‘œμ νŠΈ νšŒμ˜ν•¨ γ…‹γ…‹ μ€€λΉ„ν•œ 만큼 쒋은 ν”Όλ“œλ°± λ°›μ•„μ„œ λΏŒλ“―ν•¨ 퇴근 후에 μš΄λ™ν•˜κ³  집에 였면 μ±… 읽으며 ν•˜λ£¨ λ§ˆλ¬΄λ¦¬ν•¨

πŸ“ƒ Training data

  • korean_smile_style_dataset을 μ‚¬μš©ν•˜μ˜€μŠ΅λ‹ˆλ‹€.
  • λ³Έ λ°μ΄ν„°λŠ” Smilegate AIμ—μ„œ κ³΅κ°œν•˜λŠ” ν•œκ΅­μ–΄ 문체 μŠ€νƒ€μΌ λ³€ν™˜ "SmileStyle" λ°μ΄ν„°μ…‹μž…λ‹ˆλ‹€.
  • μ—¬λŸ¬ μŠ€νƒ€μΌμ˜ 문체가 μ‘΄μž¬ν•˜λ©° 이 μ€‘μ—μ„œ formalκ³Ό choding μŠ€νƒ€μΌμ˜ λ°μ΄ν„°λ§Œμ„ μΆ”μΆœν•˜μ—¬ μ‚¬μš©ν•˜μ˜€μŠ΅λ‹ˆλ‹€.

(The korean_smile_style_dataset was used for training. This dataset, "SmileStyle," is provided by Smilegate AI and includes various writing styles in Korean. Only the formal and elementary school styles were used in this model.)

πŸƒβ€β™‚οΈβ€βž‘οΈ Coworker

JiwookHan

Downloads last month
10
Safetensors
Model size
2.61B params
Tensor type
FP16
Β·
Inference API
Unable to determine this model's library. Check the docs .

Model tree for skwh54/Gemma-2-2b-it-elementary-style-document

Base model

google/gemma-2-2b
Finetuned
(130)
this model