app.py

import gradio as gr
from huggingface_hub import InferenceClient

# 使用可能なモデルのリスト
models = ["Sakalti/Saba1.5-Pro", "Sakalti/Saba2-Preview", "Sakalti/Saba2.1-Preview"]

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InferenceClientを初期化
    client = InferenceClient(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 = ""

    for message in client.chat_completion(
        messages,
        max_tokens=max_tokens,
        stream=True,
        temperature=temperature,
        top_p=top_p,
    ):
        token = message.choices[0].delta.content

        response += token
        yield response

# インターフェース
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Dropdown(choices=models, value=models[0], label="モデル"),
        gr.Textbox(value="ユーザーの応答と依頼に答えてください。ポジティブに", label="システムメッセージ"),
        gr.Slider(minimum=1, maximum=2048, value=512, 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)",
        ),
    ],
    concurrency_limit=30  # 例: 同時に4つのリクエストを処理
)

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