EGYADMIN/kimi-k2-thinking-dev

1import gradio as gr2import os3from huggingface_hub import InferenceClient4 5# Model configuration - Using Inference API6MODEL_NAME = "mistralai/Mistral-7B-Instruct-v0.3"7DEFAULT_SYSTEM_PROMPT = "You are a helpful AI assistant powered by Mistral."8client = None9 10def init_client():11    """Initialize the Hugging Face Inference Client"""""12    global client13    hf_token = os.environ.get("HF_TOKEN")14    if hf_token:15        client = InferenceClient(token=hf_token)16        print("Inference client initialized successfully")17        return True18    else:19        print("Warning: HF_TOKEN not found. Please set it in Space secrets.")20        return False21 22def generate_response(message, history, system_prompt, max_tokens, temperature):23    """Generate response using Hugging Face Inference API"""""24    global client25    26    if client is None:27        if not init_client():28            return "Error: HF_TOKEN not configured. Please add it in Space settings."29    30    try:31        # Build messages32        messages = [{"role": "system", "content": system_prompt or DEFAULT_SYSTEM_PROMPT}]33        34        for h in history:35            if h[0]:36                messages.append({"role": "user", "content": h[0]})37            if h[1]:38                messages.append({"role": "assistant", "content": h[1]})39        40        messages.append({"role": "user", "content": message})41        42        # Call Inference API43        response = client.chat_completion(44            model=MODEL_NAME,45            messages=messages,46            max_tokens=int(max_tokens),47            temperature=float(temperature)48        )49        50        return response.choices[0].message.content51    52    except Exception as e:53        return f"Error: {str(e)}"54 55# Create interface56print("===== Kimi K2 Thinking Dev =====")57print(f"Using Inference API with model: {MODEL_NAME}")58 59# Initialize client at startup60client_ready = init_client()61 62with gr.Blocks(title="Kimi-K2 Chat", theme=gr.themes.Soft()) as iface:63    gr.Markdown("""64    # 🤖 Kimi-K2 Instruct Chat65    **Powered by Hugging Face Inference API**66    67    This space uses the Kimi-K2-Instruct quantized model via API for efficient inference.68    """)69    70    if not client_ready:71        gr.Markdown("⚠️ **Warning:** HF_TOKEN not found. Please configure it in Space secrets.")72    73    chatbot = gr.Chatbot(height=450, label="Chat")74    75    with gr.Row():76        msg = gr.Textbox(77            placeholder="Type your message here...",78            label="Your Message",79            scale=4,80            lines=281        )82        submit_btn = gr.Button("Send 🚀", variant="primary", scale=1)83    84    with gr.Accordion("⚙️ Settings", open=False):85        system_prompt = gr.Textbox(86            value=DEFAULT_SYSTEM_PROMPT,87            label="System Prompt",88            lines=289        )90        with gr.Row():91            max_tokens = gr.Slider(92                minimum=64,93                maximum=2048,94                value=512,95                step=64,96                label="Max Tokens"97            )98            temperature = gr.Slider(99                minimum=0.1,100                maximum=2.0,101                value=0.7,102                step=0.1,103                label="Temperature"104            )105    106    clear_btn = gr.Button("🗑️ Clear Chat")107    108    def respond(message, history, system_prompt, max_tokens, temperature):109        if not message.strip():110            return "", history111        response = generate_response(message, history, system_prompt, max_tokens, temperature)112        history.append((message, response))113        return "", history114    115    msg.submit(respond, [msg, chatbot, system_prompt, max_tokens, temperature], [msg, chatbot])116    submit_btn.click(respond, [msg, chatbot, system_prompt, max_tokens, temperature], [msg, chatbot])117    clear_btn.click(lambda: [], None, chatbot)118 119if __name__ == "__main__":120    iface.launch(server_name="0.0.0.0", server_port=7860)