app.py

import gradio as gr
from huggingface_hub import InferenceClient
import time
import gc

def seize_respond(
    message,
    history,
    system_message,
    max_tokens,
    temperature,
    top_p,
    hf_token,
):
    """
    🎯 Seize: Direct Serious AI powered by efficient model
    """
    if not hf_token:
        yield "Error: No Hugging Face token provided. Please log in."
        return
    
    # Use smaller, more reliable model
    client = InferenceClient(
        token=hf_token, 
        model="HuggingFaceH4/zephyr-7b-beta"
    )
    
    # Limit conversation history to prevent memory issues
    if len(history) > 6:
        history = history[-6:]
    
    # Seize2B system prompt
    seize2b_system = """You are Seize, a direct AI assistant.

IDENTITY: Seize (Efficient Model)
STYLE: Direct, serious, professional
RULES:
1. Respond with maximum conciseness
2. No greetings, emojis, or casual phrases
3. If question unclear, request clarification directly
4. Use bullet points for lists
5. Never apologize or say "I'm here to help"
6. Focus only on the question asked
7. If you don't know, say "No data" or "Cannot compute"
8. Default response length: 1-3 sentences

EXAMPLE INTERACTIONS:
User: what's good
Seize: Systems operational. Inquiry?

User: explain quantum computing
Seize: Quantum computing: Qubits, superposition, entanglement. Enables parallel processing."""
    
    messages = [{"role": "system", "content": seize2b_system}]
    
    # Convert Gradio history format to chat completion format
    for h in history:
        messages.append({"role": "user", "content": h[0]})
        messages.append({"role": "assistant", "content": h[1]})
    
    messages.append({"role": "user", "content": message})
    
    response = ""
    
    # Rate limiting delay
    time.sleep(0.3)
    
    try:
        stream = client.chat_completion(
            messages,
            max_tokens=min(max_tokens, 256),
            stream=True,
            temperature=min(temperature, 0.7),
            top_p=top_p,
        )
        
        for chunk in stream:
            if chunk.choices and chunk.choices[0].delta.content:
                token = chunk.choices[0].delta.content
                response += token
                time.sleep(0.005)
                yield response
                
    except Exception as e:
        error_msg = str(e)
        if "rate limit" in error_msg.lower():
            yield "Rate limit exceeded. Please wait 10-20 seconds before next message."
        elif "memory" in error_msg.lower() or "out of memory" in error_msg.lower():
            yield "Memory limit reached. Please use shorter prompts or wait a moment."
        elif "authentication" in error_msg.lower():
            yield "Authentication failed. Please check your Hugging Face token."
        else:
            yield f"Error: {error_msg[:80]}"
    
    # Cleanup
    gc.collect()

# Create the chatbot with all parameters
chatbot = gr.ChatInterface(
    fn=seize_respond,
    additional_inputs=[
        gr.Textbox("You are Seize. Be direct and serious.", label="System Message", visible=False),
        gr.Slider(64, 512, value=256, step=64, label="Max Tokens"),
        gr.Slider(0.1, 1.0, value=0.3, step=0.1, label="Temperature"),
        gr.Slider(0.1, 1.0, value=0.9, step=0.1, label="Top-p"),
        gr.Textbox(label="HF Token", type="password", visible=False)
    ],
    type="messages",
    title="Seize (Efficient Model)",
    description="Direct AI assistant. No fluff. Serious responses only.",
    theme=gr.themes.Soft(
        primary_hue="blue",
        secondary_hue="gray",
        neutral_hue="gray",
    ),
    examples=[
        ["what's good"],
        ["Explain quantum computing"],
        ["List 3 benefits of exercise"],
        ["Python fibonacci function"],
    ],
    cache_examples=False,
    retry_btn=None,
    undo_btn=None,
    clear_btn="🗑️ Clear",
    submit_btn="🚀 Ask Seize",
)

# Create the full interface
with gr.Blocks(css="""
    .seize-header {
        text-align: center;
        padding: 20px;
        background: linear-gradient(135deg, #1e3c72 0%, #2a5298 100%);
        color: white;
        border-radius: 10px;
        margin-bottom: 20px;
    }
    .seize-title {
        font-size: 2.5em;
        font-weight: 800;
        margin: 0;
        letter-spacing: -0.5px;
    }
    .seize-subtitle {
        font-size: 1.1em;
        opacity: 0.9;
        margin-top: 5px;
    }
    .seize-footer {
        text-align: center;
        padding: 15px;
        font-size: 0.9em;
        color: #666;
        border-top: 1px solid #e0e0e0;
        margin-top: 20px;
    }
    .gradio-container {
        max-width: 800px !important;
        margin: 0 auto !important;
    }
""") as demo:
    
    # Seize Header
    gr.HTML("""
    <div class="seize-header">
        <h1 class="seize-title">SEIZE</h1>
        <p class="seize-subtitle">Direct AI Assistant • Efficient Model</p>
        <p style="font-size: 0.9em; margin-top: 10px; opacity: 0.8;">
            Serious responses. No fluff. Optimized for stability.
        </p>
    </div>
    """)
    
    # Authentication notice
    gr.Markdown("""
    ### ⚠️ Authentication Required
    You need a Hugging Face token to use this model. Get one at [huggingface.co/settings/tokens](https://huggingface.co/settings/tokens)
    
    **For testing without token:** Try the examples below first.
    """)
    
    # Render the chatbot
    chatbot.render()
    
    # Settings in accordion
    with gr.Accordion("⚙️ Advanced Settings", open=False):
        gr.Markdown("""
        - **Max Tokens**: Controls response length (64-512)
        - **Temperature**: Higher = more creative, Lower = more focused
        - **Top-p**: Controls vocabulary diversity
        """)
    
    # Seize Footer
    gr.HTML("""
    <div class="seize-footer">
        <strong>Optimized for Hugging Face Spaces</strong> • Model: zephyr-7b-beta
        <br>
        <small>Response time: ~1-3s • Max tokens: 256 • Temperature: 0.3</small>
    </div>
    """)

# Launch configuration
if __name__ == "__main__":
    demo.launch(
        server_name="0.0.0.0",
        server_port=7860,
        show_error=True,
        debug=False,
        share=False  # Set to True for temporary public link
    )