import gradio as gr
from huggingface_hub import InferenceClient
import os

# Deutsche LLM Konfiguration
HF_TOKEN = os.getenv("tomoniaccess")
current_model = "LeoLM/leo-hessianai-13b-chat"
client = InferenceClient(model=current_model, token=HF_TOKEN)

conversation_history = []

def enhanced_chat_response(user_input, max_tokens, temperature, top_p):
    if not user_input.strip():
        return "", "Bitte gib eine Nachricht ein.", "", ""
    
    messages = [{"role": "user", "content": user_input}]
    
    response_text = ""
    try:
        for message in client.chat_completion(
            messages=messages,
            max_tokens=min(max_tokens, 100),
            stream=True,
            temperature=temperature,
            top_p=top_p
        ):
            token = message.choices[0].delta.content
            if token:
                response_text += token

    except Exception as e:
        print(f"API Error: {e}")  # This will show in console
        print(f"Error type: {type(e)}")
        response_text = f"API Fehler: {str(e)}"  # Show actual error to user
    
    
    response_text = response_text.strip()
    
    chat_display = f"**Du:** {user_input}\n**Assistant:** {response_text}\n\n"
    
    return "", response_text, chat_display, ""

def reset_conversation():
    return "Neues Gespräch gestartet.", "", ""

with gr.Blocks(title="Depression Training Simulator", theme=gr.themes.Soft()) as demo:
    gr.Markdown("# 🧠 Depression Training Simulator")
    gr.Markdown("**Übe realistische Gespräche mit depressiven Jugendlichen und erhalte Feedback**")
    
    with gr.Row():
        with gr.Column(scale=1):
            # Parameter
            gr.Markdown("### ⚙️ Einstellungen")
            max_tokens = gr.Slider(50, 150, value=80, step=10, label="Antwortlänge")
            temperature = gr.Slider(0.5, 1.2, value=0.9, step=0.1, label="Variabilität")
            top_p = gr.Slider(0.7, 1.0, value=0.95, step=0.05, label="Fokus")
            
            # Actions
            gr.Markdown("### 🔄 Aktionen")
            reset_btn = gr.Button("Neues Gespräch", variant="secondary")
            
        with gr.Column(scale=2):
            # Chat Interface
            gr.Markdown("### 💬 Gespräch")
            
            user_input = gr.Textbox(
                label="Deine Nachricht", 
                placeholder="Beginne das Gespräch...",
                lines=2
            )
            send_btn = gr.Button("📨 Senden", variant="primary")
            
            bot_response = gr.Textbox(
                label="Antwort", 
                interactive=False,
                lines=3
            )
            
            chat_history = gr.Textbox(
                label="Gesprächsverlauf", 
                interactive=False,
                lines=12
            )
            
            # Feedback Panel (empty, no persona feedback)
            with gr.Accordion("📈 Trainer-Feedback", open=True):
                feedback_display = gr.Markdown("Starte ein Gespräch, um Feedback zu erhalten.")
    
    # Event Bindings
    send_btn.click(
        fn=enhanced_chat_response,
        inputs=[user_input, max_tokens, temperature, top_p],
        outputs=[user_input, bot_response, chat_history, feedback_display]
    )
    
    user_input.submit(
        fn=enhanced_chat_response,
        inputs=[user_input, max_tokens, temperature, top_p],
        outputs=[user_input, bot_response, chat_history, feedback_display]
    )
    
    reset_btn.click(
        fn=reset_conversation,
        outputs=[bot_response, chat_history, feedback_display]
    )

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