File size: 3,995 Bytes
8fce01d be423b2 8fce01d be423b2 20c3afe be423b2 20c3afe be423b2 20c3afe be423b2 20c3afe be423b2 20c3afe be423b2 20c3afe be423b2 |
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 |
---
license: apache-2.0
datasets:
- brian-lim/smile_style_orca
language:
- ko
---
# Korean Style Transfer
This model is a fine-tuned version of [Synatra-7B-v0.3-dpo](https://huggingface.co/maywell/Synatra-7B-v0.3-dpo) using a Korean style dataset provided by Smilegate AI (https://github.com/smilegate-ai/korean_smile_style_dataset/tree/main).
Since the original dataset is tabular and not fit for training the LLM, I have preprocessed it into an instruction-input-output format, which can be found [here](https://huggingface.co/datasets/brian-lim/smile_style_orca).
The dataset is then fed into the ChatML template. Feel free to use my version of the dataset as needed.
ν΄λΉ λͺ¨λΈμ [Synatra-7B-v0.3-dpo](https://huggingface.co/maywell/Synatra-7B-v0.3-dpo) λͺ¨λΈμ μ€λ§μΌκ²μ΄νΈ AIμμ μ 곡νλ Smile style λ°μ΄ν°μ
μΌλ‘ νμΈνλ νμ΅λλ€.
κΈ°μ‘΄ λ°μ΄ν°μ
μ ν
μ΄λΈ ννλ‘ λμ΄μμ΄ ν΄λΉ λ°μ΄ν°λ₯Ό instruction-input-output ννλ‘ λ§λ€μκ³ , [μ¬κΈ°](https://huggingface.co/datasets/brian-lim/smile_style_orca)μμ νμΈ κ°λ₯ν©λλ€.
λ°μ΄ν°μ
μ λΆλ¬μ¨ λ€ ChatML νμμ λ§μΆ° νλ ¨ λ°μ΄ν° ꡬμΆμ ν λ€ μ§ννμ΅λλ€. νμνμλ€λ©΄ μμ λ‘κ² μ¬μ©νμκΈ° λ°λλλ€.
# How to use
```python
>>> import torch
>>> from transformers import AutoModelForCausalLM, AutoTokenizer
device = 'cuda' if torch.cuda.is_available() else 'cpu'
tokenizer = AutoTokenizer.from_pretrained('brian-lim/smile-style-transfer')
model = AutoModelForCausalLM.from_pretrained('brian-lim/smile-style-transfer', device_map = device)
prompts = {'informal': 'μ£Όμ΄μ§ κΈμ κ°λ₯ν νμμ μ΄μ§ μκ³ λ±λ±νμ§ μμ λνμ²΄λ‘ λ°κΏμ€.',
'android': 'μ£Όμ΄μ§ κΈμ κ°λ₯ν μλλ‘μ΄λ λ‘λ΄κ³Ό κ°μ λνμ²΄λ‘ λ°κΏμ€.',
'azae': 'μ£Όμ΄μ§ κΈμ κ°λ₯ν μμ μ¨κ°μ λ§ν¬λ‘ λ°κΏμ€.',
'chat': 'μ£Όμ΄μ§ κΈμ κ°λ₯ν μΈν°λ·μμ μ¬μ©νλ λ§ν¬λ‘ λ°κΏμ€.',
'choding': 'μ£Όμ΄μ§ κΈμ κ°λ₯ν μ΄λ±νμμ²λΌ μ§§κ² μ€μΈ λνμ²΄λ‘ λ°κΏμ€.',
'emoticon': 'μ£Όμ΄μ§ κΈμ κ°λ₯ν μ΄λͺ¨ν°μ½μ΄ λ€μ΄κ° λνμ²΄λ‘ λ°κΏμ€.',
'enfp': 'μ£Όμ΄μ§ κΈμ κ°λ₯ν νκΈ°μ°¨λ©΄μ 곡κ°μ λ§μ΄ νλ μΉμ ν λνμ²΄λ‘ λ°κΏμ€.',
'gentle' : 'μ£Όμ΄μ§ κΈμ κ°λ₯ν βμβλ‘ λλμ§ μμΌλ©΄μ κΉλν λνμ²΄λ‘ λ°κΏμ€.',
'halbae' : 'μ£Όμ΄μ§ κΈμ κ°λ₯ν μ°λ₯μ΄ μλ ν μλ²μ§ κ°μ 맑ν¬λ‘ λ°κΏμ€.',
'halmae' : 'μ£Όμ΄μ§ κΈμ κ°λ₯ν λΉμμ΄κ° λ€μ΄κ°λ ν λ¨Έλ κ°μ 맑ν¬λ‘ λ°κΏμ€.',
'joongding': 'μ£Όμ΄μ§ κΈμ κ°λ₯ν μ€νκ΅ 2νλ
μ λ§ν¬λ‘ λ°κΏμ€.',
'king': 'μ£Όμ΄μ§ κΈμ κ°λ₯ν μ‘°μ μλ μμ λ§ν¬λ‘ λ°κΏμ€.',
'seonbi': 'μ£Όμ΄μ§ κΈμ κ°λ₯ν μ‘°μ μλ μ λΉμ λ§ν¬λ‘ λ°κΏμ€.',
'sosim': 'μ£Όμ΄μ§ κΈμ κ°λ₯ν μμ£Ό μμ¬νκ³ μ‘°μ¬μ€λ¬μ΄ λ§ν¬λ‘ λ°κΏμ€.',
'translator': 'μ£Όμ΄μ§ κΈμ κ°λ₯ν μ΄μν νκ΅μ΄ λ²μ λ§ν¬λ‘ λ°κΏμ€.',
}
query = '[INPUT]: μλ
νμΈμ. μμ¦ λ μ¨κ° λ§μ΄ μμνλ€μ \n[OUTPUT]: '
input_query = prompts['king'] + query
input_tokenized = tokenizer(input_query,return_tensors="pt").to(device)
g_config = GenerationConfig(temperature=0.3,
repetition_penalty=1.2,
max_new_tokens=768,
do_sample=True,
)
output = model.generate(**input_tokenized,
generation_config=g_config,
pad_token_id=tokenizer.eos_token_id,
eos_token_id=tokenizer.eos_token_id,)
output_text = tokenizer.decode(output.detach().cpu().numpy()[0])
output_text = output_text[output_text.find('[OUTPUT]'):]
print(output_text)
```
---
license: apache-2.0
--- |