Instructions to use esomnmms/finetuning-lora with libraries, inference providers, notebooks, and local apps. Follow these links to get started.
- Libraries
- PEFT
How to use esomnmms/finetuning-lora with PEFT:
from peft import PeftModel from transformers import AutoModelForCausalLM base_model = AutoModelForCausalLM.from_pretrained("unsloth/gemma-3-4b-it-unsloth-bnb-4bit") model = PeftModel.from_pretrained(base_model, "esomnmms/finetuning-lora") - Notebooks
- Google Colab
- Kaggle
- Local Apps Settings
- Unsloth Desktop
finetuning-lora
unsloth/gemma-3-4b-it๋ฅผ LoRA๋ก ํ์ธํ๋ํ ์ด๋ํฐ์
๋๋ค. ํ๊ตญ์ด ์บ๋ฆญํฐ(๋งํฌ/ํค) ๋กคํ๋ ์ด ๋ํ ๋ฐ์ดํฐ๋ก SFT๋ฅผ ์งํํด, ์ง์ ํ ์บ๋ฆญํฐ์ ๋งํฌ๋ก ๋ตํ๋๋ก ๋ง๋ค์์ต๋๋ค.
ํ์ต ์ค์
| ํญ๋ชฉ | ๊ฐ |
|---|---|
| Base model | unsloth/gemma-3-4b-it (4bit ์์ํ ๋ก๋) |
| ๋ฐฉ์ | LoRA (r=8, alpha=8), ์ธ์ด ๋ชจ๋ธ ๋ ์ด์ด๋ง ํ์ต |
| ํ๋ ์์ํฌ | Unsloth + TRL (SFT) |
| ๋ฐ์ดํฐ | huggingface-KREW/korean-role-playing์ exa-data (Apache 2.0, ํ๊ตญ์ด). 512ํ ํฐ ์ดํ 888๊ฐ๋ฅผ train 710 / valid 89 / test 89๋ก ๋ถํ |
์ฌ์ฉ ๋ฐฉ๋ฒ
from unsloth import FastModel
model, processor = FastModel.from_pretrained(
model_name="esomnmms/finetuning-lora",
load_in_4bit=True,
)
๋ผ์ด์ ์ค
Base model์ธ Gemma 3์ Gemma Terms of Use๋ฅผ ๋ฐ๋ฆ ๋๋ค.
์ฝ๋
ํ์ต ๋ ธํธ๋ถ๊ณผ ๋ฐ์ดํฐ ์ฒ๋ฆฌ ์ฝ๋: https://github.com/mnmms/sojeong-finetuning
- Downloads last month
- 10