app.py

import gradio as gr
from modelscope.pipelines import pipeline
from modelscope.utils.constant import Tasks

# 使用可能なモデルのリスト
models = ["Qwen/Qwen2.5-0.5B-Instruct", "Qwen/Qwen2.5-1.5B-Instruct"]

# モデルを事前にロード
model_pipelines = {model: pipeline(task=Tasks.text_generation, model=model) for model in models}

def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
    selected_model
):
    # 選択したモデルに基づいてPipelineを選択
    pipe = model_pipelines[selected_model]

    messages = [{"role": "system", "content": system_message}]

    for val in history:
        if val[0]:
            messages.append({"role": "user", "content": val[0]})
        if val[1]:
            messages.append({"role": "assistant", "content": val[1]})

    messages.append({"role": "user", "content": message})

    response = ""

    # モデルの推論
    for message in pipe(messages, max_tokens=max_tokens, temperature=temperature, top_p=top_p, stream=True):
        token = message.get("choices", [{}])[0].get("delta", {}).get("content", "")
        response += token

    return response

# インターフェース
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="あなたはフレンドリーなチャットボットです。", label="システムメッセージ"),
        gr.Slider(minimum=1, maximum=2048, value=768, step=1, label="新規トークン最大"),
        gr.Slider(minimum=0.1, maximum=4.0, value=0.7, step=0.1, label="温度"),
        gr.Slider(minimum=0.1, maximum=1.0, value=0.95, step=0.05, label="Top-p (核 sampling)"),
        gr.Dropdown(choices=models, value=models[0], label="モデル"),
    ],
    concurrency_limit=30  # 例: 同時に30つのリクエストを処理
)

if __name__ == "__main__":
    demo.launch(share=True)