Housing-Subscription-QA-Phi-3.5

Housing-Subscription-QA-Phi-3.5

<!-- Provide a quick summary of what the model is/does. -->

Model Details

Model Description

<!-- Provide a longer summary of what this model is. -->

• Model type: Question Answering
• Language(s) (NLP): Korean
• Finetuned from model: microsoft/Phi-3.5-mini-instruct

Model Sources

from peft import PeftModel, PeftConfig
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline

config = PeftConfig.from_pretrained("hecatonai/Housing-Subscription-QA-Phi-3.5")
base_model = AutoModelForCausalLM.from_pretrained("microsoft/Phi-3.5-mini-instruct", device_map='auto')
model = PeftModel.from_pretrained(base_model, "hecatonai/Housing-Subscription-QA-Phi-3.5", device_map='auto')

# 토크나이저 로드
tokenizer = AutoTokenizer.from_pretrained("microsoft/Phi-3.5-mini-instruct", device_map='auto')

# 입력 텍스트 포맷팅
def apply_chat_template(question):
    template = "<|system|>\nYou are a helpful AI assistant. The default is 2024.<|end|>\n<|user|>\n{question}<|end|>\n<|assistant|>\n"
    return template.format(question=question)

# 입력 텍스트 토크나이징
question = "투기과열지구 또는 청약과열지역에서 외국인 1순위 청약 가능?"
input_text = apply_chat_template(question)

inputs = tokenizer(input_text, return_tensors="pt")

# 예측 수행
outputs = model.generate(**inputs, max_length=1000)

# 출력 디코딩
decoded_output = tokenizer.decode(outputs[0], skip_special_tokens=True)
print(decoded_output)

Bias, Risks, and Limitations

해당 모델은 대한민국 국토교통부에서 발행한 2022년도 및 2024년도 주택청약 FAQ를 기반으로
Fine-Tune 한 LLM입니다. 따라서 해당 FAQ에 포함되지 않은 질문에 대해서는 부정확한 답변을 할 수 있으니 사용에 유의바랍니다.

How to Get Started with the Model

Use the code below to get started with the model.

from peft import PeftModel
from transformers import AutoModelForCausalLM, AutoTokenizer

config = PeftConfig.from_pretrained("hecatonai/Housing-Subscription-QA-Phi-3.5")
base_model = AutoModelForCausalLM.from_pretrained("microsoft/Phi-3.5-mini-instruct", device_map='auto')
model = PeftModel.from_pretrained(base_model, "hecatonai/Housing-Subscription-QA-Phi-3.5", device_map='auto')

Using with Pipeline

import torch
from transformers import AutoModelForCausalLM, AutoTokenizer, pipeline

model = AutoModelForCausalLM.from_pretrained("hecatonai/Housing-Subscription-QA-Phi-3.5", device_map='auto')

pipe = pipeline("text-generation", model=model, tokenizer="microsoft/Phi-3.5-mini-instruct", torch_dtype=torch.bfloat16, device_map="auto")
messages = [
    {"role": "system", "content": "You are a helpful AI assistant.  The default is 2024."},
    {"role": "user", "content": "투기과열지구 및 청약과열지역 1순위 제한대상 누구?"}
]

prompt = pipe.tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
outputs = pipe(prompt, renormalize_logits=True, max_new_tokens=512, do_sample=False)
print(outputs[0]["generated_text"])

Result

<|system|>
You are a helpful AI assistant.  The default is 2024.<|end|>
<|user|>
투기과열지구 및 청약과열지역 1순위 제한대상 누구?<|end|>
<|assistant|>
2024년 답변: 투기과열지구 및 청약과열지역에서 국민주택과 민영주택 1순위 제한 대상은, 과거 5년 이내에 본인 또는 세대원이 다른 주택의 당첨자가 된 경우입니다.

Training Details

Training Data

<!-- This should link to a Dataset Card, perhaps with a short stub of information on what the training data is all about as well as documentation related to data pre-processing or additional filtering. -->

dataset: Housing_Subscription_QA_Dataset

Training Hyperparameters

This model following Hyperparameters were used during training:

• bf16 = True
• learning_rate = 5.0e-5
• num_train_epochs = 15
• per_device_batch_size = 4
• warmup_ratio = 0.2

Traning Prompt

 messages = [{"role": "system", "content": "You are a helpful AI assistant."},
    {"role": "user", "content": f"{example['question']}"},
    {"role": "assistant", "content": f"{example['answer']}"}]

Framework versions

• PEFT 0.12.0
• Transformers 4.44.2