import gradio as gr import requests import os import json from collections import deque import asyncio import aiohttp TOKEN = os.getenv("HUGGINGFACE_API_TOKEN") if not TOKEN: raise ValueError("API token is not set. Please set the HUGGINGFACE_API_TOKEN environment variable.") memory = deque(maxlen=10) async def respond( message, history: list[tuple[str, str]], system_message="AI Assistant Role", max_tokens=512, temperature=0.7, top_p=0.95, ): system_prefix = "System: 입력어의 언어(영어, 한국어, 중국어, 일본어 등)에 따라 동일한 언어로 답변하라." full_system_message = f"{system_prefix}{system_message}" memory.append((message, None)) messages = [{"role": "system", "content": full_system_message}] for val in memory: if val[0]: messages.append({"role": "user", "content": val[0]}) if val[1]: messages.append({"role": "assistant", "content": val[1]}) headers = { "Authorization": f"Bearer {TOKEN}", "Content-Type": "application/json" } payload = { "model": "meta-llama/Meta-Llama-3.1-70B-Instruct", "max_tokens": max_tokens, "temperature": temperature, "top_p": top_p, "messages": messages, "stream": True } async with aiohttp.ClientSession() as session: async with session.post("https://api-inference.huggingface.co/v1/chat/completions", headers=headers, json=payload) as response: partial_words = "" async for chunk in response.content: if chunk: chunk_data = chunk.decode('utf-8') if chunk_data.startswith("data: "): chunk_data = chunk_data[6:] try: response_json = json.loads(chunk_data) if "choices" in response_json: delta = response_json["choices"][0].get("delta", {}) if "content" in delta: content = delta["content"] partial_words += content yield partial_words except json.JSONDecodeError: continue theme = "Nymbo/Nymbo_Theme" css = """ footer { visibility: hidden; } """ demo = gr.ChatInterface( css=css, fn=respond, theme=theme, additional_inputs=[ gr.Textbox(value="AI Assistant Role", label="System message"), gr.Slider(minimum=1, maximum=2048, value=512, step=1, label="Max new tokens"), gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="Temperature"), gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (nucleus sampling)"), ] ) if __name__ == "__main__": demo.queue().launch(max_threads=20)