import gradio as gr from huggingface_hub import InferenceClient import os import random import logging # 로깅 설정 logging.basicConfig(filename='language_model_playground.log', level=logging.DEBUG, format='%(asctime)s - %(levelname)s - %(message)s') # 모델 목록 MODELS = { "Zephyr 7B Beta": "HuggingFaceH4/zephyr-7b-beta", "DeepSeek Coder V2": "deepseek-ai/DeepSeek-Coder-V2-Instruct", "Meta Llama 3.1 8B": "meta-llama/Meta-Llama-3.1-8B-Instruct", "Meta-Llama 3.1 70B-Instruct": "meta-llama/Meta-Llama-3.1-70B-Instruct", "Microsoft": "microsoft/Phi-3-mini-4k-instruct", "Mixtral 8x7B": "mistralai/Mistral-7B-Instruct-v0.3", "Mixtral Nous-Hermes": "NousResearch/Nous-Hermes-2-Mixtral-8x7B-DPO", "Cohere Command R+": "CohereForAI/c4ai-command-r-plus", "Aya-23-35B": "CohereForAI/aya-23-35B" } # HuggingFace 토큰 설정 hf_token = os.getenv("HF_TOKEN") if not hf_token: raise ValueError("HF_TOKEN 환경 변수가 설정되지 않았습니다.") def call_hf_api(prompt, reference_text, max_tokens, temperature, top_p, model): client = InferenceClient(model=model, token=hf_token) combined_prompt = f"{prompt}\n\n참고 텍스트:\n{reference_text}" random_seed = random.randint(0, 1000000) try: response = client.text_generation( combined_prompt, max_new_tokens=max_tokens, temperature=temperature, top_p=top_p, seed=random_seed ) return response except Exception as e: logging.error(f"HuggingFace API 호출 중 오류 발생: {str(e)}") return f"응답 생성 중 오류 발생: {str(e)}. 나중에 다시 시도해 주세요." def generate_response(prompt, reference_text, max_tokens, temperature, top_p, model): response = call_hf_api(prompt, reference_text, max_tokens, temperature, top_p, MODELS[model]) response_html = f"""

생성된 응답:

{response}
""" return response_html # Gradio 인터페이스 설정 with gr.Blocks() as demo: gr.Markdown("## 언어 모델 프롬프트 플레이그라운드") with gr.Column(): model_radio = gr.Radio(choices=list(MODELS.keys()), value="Zephyr 7B Beta", label="언어 모델 선택") prompt_input = gr.Textbox(label="프롬프트 입력", lines=5) reference_text_input = gr.Textbox(label="참고 텍스트 입력", lines=5) with gr.Row(): max_tokens_slider = gr.Slider(minimum=0, maximum=5000, value=2000, step=100, label="최대 토큰 수") temperature_slider = gr.Slider(minimum=0, maximum=1, value=0.75, step=0.05, label="온도") top_p_slider = gr.Slider(minimum=0, maximum=1, value=0.95, step=0.05, label="Top P") generate_button = gr.Button("응답 생성") response_output = gr.HTML(label="생성된 응답") # 버튼 클릭 시 응답 생성 generate_button.click( generate_response, inputs=[prompt_input, reference_text_input, max_tokens_slider, temperature_slider, top_p_slider, model_radio], outputs=response_output ) # 인터페이스 실행 demo.launch(share=True)