app.py

import gradio as gr
from huggingface_hub import InferenceClient

# ★ モデルを gemma-3-27b-it-abliterated に変更
#    provider="hf-inference" でHugging Face Inference APIを明示的に指定
client = InferenceClient(
    model="mlabonne/gemma-3-12b-it-abliterated"
)  # :contentReference[oaicite:0]{index=0}

def respond(
    message,
    history: list[tuple[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
):
    # system_message→history→最新ユーザー発話 の順に messages を構築
    messages = [{"role": "system", "content": system_message}]
    for u, a in history:
        if u:
            messages.append({"role": "user", "content": u})
        if a:
            messages.append({"role": "assistant", "content": a})
    messages.append({"role": "user", "content": message})

    # chat_completion を呼び出し（stream=True でトークン毎に返す）
    response = ""
    for chunk in client.text_to_text(
        messages=messages,
        max_tokens=max_tokens,
        temperature=temperature,
        top_p=top_p,
        stream=True,
    ):
        delta = chunk.choices[0].delta.content
        response += delta
        yield response

# GradioのチャットUIをそのまま利用
demo = gr.ChatInterface(
    respond,
    additional_inputs=[
        gr.Textbox(value="You are a friendly Chatbot.", 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.launch()