ChatSKKU5.8B / README.md
jojo0217's picture
Update README.md
bb273c4
|
raw
history blame
1.65 kB
---
license: apache-2.0
datasets:
- jojo0217/korean_rlhf_dataset
language:
- ko
---
μ„±κ· κ΄€λŒ€ν•™κ΅ μ‚°ν•™ν˜‘λ ₯ κ³Όμ •μ—μ„œ λ§Œλ“  ν…ŒμŠ€νŠΈ λͺ¨λΈμž…λ‹ˆλ‹€.
ν•™μŠ΅ λ°μ΄ν„°μ˜ μ°Έκ³  λͺ¨λΈμ΄λΌκ³  μƒκ°ν•˜μ‹œλ©΄ 쒋을 것 κ°™μŠ΅λ‹ˆλ‹€.
κΈ°μ‘΄ 10만 7천개의 데이터 + 2천개 μΌμƒλŒ€ν™” μΆ”κ°€ 데이터λ₯Ό μ²¨κ°€ν•˜μ—¬ ν•™μŠ΅ν•˜μ˜€μŠ΅λ‹ˆλ‹€.
μΈ‘μ •ν•œ kobest μ μˆ˜λŠ” λ‹€μŒκ³Ό κ°™μŠ΅λ‹ˆλ‹€.
![score](./asset/score.png)
λͺ¨λΈ prompt templateλŠ” kullm의 templateλ₯Ό μ‚¬μš©ν•˜μ˜€μŠ΅λ‹ˆλ‹€.
ν…ŒμŠ€νŠΈ μ½”λ“œλŠ” λ‹€μŒκ³Ό κ°™μŠ΅λ‹ˆλ‹€.
```
from transformers import pipeline, AutoModelForCausalLM, AutoTokenizer
model_name="jojo0217/ChatSKKU5.8B"
tokenizer = AutoTokenizer.from_pretrained(model_name)
model = AutoModelForCausalLM.from_pretrained(
model_name,
device_map="auto",
load_in_8bit=True,#λ§Œμ•½ μ–‘μžν™” 끄고 μ‹Άλ‹€λ©΄ false
)
tokenizer = AutoTokenizer.from_pretrained(model_name)
pipe = pipeline(
"text-generation",
model=model,
tokenizer=model_name,
device_map="auto"
)
def answer(message):
prompt=f"μ•„λž˜λŠ” μž‘μ—…μ„ μ„€λͺ…ν•˜λŠ” λͺ…λ Ήμ–΄μž…λ‹ˆλ‹€. μš”μ²­μ„ 적절히 μ™„λ£Œν•˜λŠ” 응닡을 μž‘μ„±ν•˜μ„Έμš”.\n\n### λͺ…λ Ήμ–΄:\n{message}"
ans = pipe(
prompt + "\n\n### 응닡:",
do_sample=True,
max_new_tokens=512,
temperature=0.9,
num_beams = 1,
repetition_penalty = 1.0,
return_full_text=False,
eos_token_id=2,
)
msg = ans[0]["generated_text"]
return msg
answer('μ„±κ· κ΄€λŒ€ν•™κ΅μ—λŒ€ν•΄ μ•Œλ €μ€˜')
```