- Developed by: lwef
- License: apache-2.0
- Finetuned from model : beomi/Llama-3-Open-Ko-8B
korean dialogue summary fine-tuned model
how to use
prompt_template = '''
μλ λνλ₯Ό μμ½ν΄ μ£ΌμΈμ. λν νμμ '#λν μ°Έμ¬μ#: λν λ΄μ©'μ
λλ€.
### λν >>>{dialogue}
### μμ½ >>>'''
if True:
from unsloth import FastLanguageModel
model, tokenizer = FastLanguageModel.from_pretrained(
model_name = "lwef/llama3-8B-ko-dialogue-summary-finetuned", # YOUR MODEL YOU USED FOR TRAINING
max_seq_length = 2048,
dtype = None,
load_in_4bit = True,
)
FastLanguageModel.for_inference(model) # Enable native 2x faster inference
dialogue = '''#P01#: μ νμΆ κ³Όμ λ무 μ΄λ €μ... 5μͺ½ μΈκ² μλλ° γ
‘γ
‘ #P02#: λͺ¬λλͺ¬λλκ°λμμ¨ γ
γ
#P01#: 5μͺ½ λμΆ© μμμ νλ¦λλ‘ μ μ¨μΌμ§..μ΄μ 1μͺ½μ ;; 5μͺ½ μλ λ€μ€λ§ μ μ΄μΌμ§ #P02#: μλ... λκ°λΆλμ€μν κ±°κ°μ κ±°μκ½μ±μμμ°μ
#P01#: λͺ»μ¨ μΈλ§μ
μ¨ #P02#: μ΄κ±°μ€κ°λ체μ¬?? #P01#: γ΄γ΄ κ·Έλ₯ κ³Όμ μ κ·Έλμ λ μ§μ¦λ¨'''
formatted_prompt = prompt_template.format(dialogue=dialogue)
# ν ν¬λμ΄μ§
inputs = tokenizer(
formatted_prompt,
return_tensors="pt"
).to("cuda")
outputs = model.generate(
**inputs,
max_new_tokens = 128,
eos_token_id=tokenizer.eos_token_id, # EOS ν ν°μ μ¬μ©νμ¬ λͺ
μμ μΌλ‘ μΆλ ₯μ λμ μ§μ .
use_cache = True
)
decoded_outputs = tokenizer.batch_decode(outputs, skip_special_tokens=True)
result = decoded_outputs[0]
print(result)
result = result.split('### μμ½ >>>')[-1].strip()
print(result)
This llama model was trained 2x faster with Unsloth and Huggingface's TRL library.
Model tree for lwef/llama3-8B-ko-dialogue-summary-finetuned
Base model
beomi/Llama-3-Open-Ko-8B