from openboxing_api import get_all_champions, find_champion_by_name, get_all_bouts, get_bouts_for_champion
import gradio as gr
from huggingface_hub import InferenceClient


# -------------------------------------------------------
# SYSTEM PROMPT (Upgraded for DeepSeek Reasoning)
# -------------------------------------------------------
system_prompt = """
You are BOXTRON-AI, an elite boxing analyst and professional fight judge. 
Your expertise includes stylistic breakdowns, matchup analysis, round-by-round simulations, 
and probability-based fight predictions. 

Your responsibilities:
- Score rounds using the official 10-point must system.
- Provide deep, technical, and objective fight analysis.
- Break down styles, strengths, weaknesses, pace, footwork, combinations, IQ, and strategy.
- Analyze attributes such as height, reach, stance, age, KO %, defense, accuracy, and habits.
- Predict multiple realistic scenarios, assign probabilities, and justify them.
- For simulations, describe round-by-round action and scoring.
- Clearly explain uncertainty when present.
- NEVER invent fake fight records or medical information.
- Base predictions on logic, style interaction, known tendencies, and probability.

Your tone:
Professional, analytical, neutral, expert-level, similar to a mix of a ringside commentator 
and a veteran boxing judge.

Whenever the user mentions two fighters, assume they want:
1. A matchup breakdown
2. A stylistic analysis
3. A round-by-round prediction (if asked)
4. Probabilities for each plausible outcome

Always think step-by-step and use reasoning before concluding.
"""


# -------------------------------------------------------
# CHAT RESPONSE FUNCTION
# -------------------------------------------------------
def respond(
    message,
    history: list[dict[str, str]],
    system_message,
    max_tokens,
    temperature,
    top_p,
    hf_token: gr.OAuthToken,
):
    # Example prompt: predict shakur stevenson vs teofimo lopez
    if "predict" in message.lower() and "vs" in message.lower():
        message = message.replace("predict ", "")
        print(message)
        fighters = message.split(" vs ")
        print(fighters)
        if len(fighters) == 2:
            all_champs = get_all_champions()
            all_bouts = get_all_bouts()
            
            fighter1_name = fighters[0].strip()
            fighter2_name = fighters[1].strip()
    
            fighter1 = find_champion_by_name(all_champs, fighter1_name)
            fighter2 = find_champion_by_name(all_champs, fighter2_name)
    
            if fighter1 and fighter2:
                history1 = get_bouts_for_champion(all_bouts, fighter1["championId"])
                history2 = get_bouts_for_champion(all_bouts, fighter2["championId"])
    
                # Build structured prompt for LLM
                message = f"""
                Fighter 1: {fighter1['name']['first']} {fighter1['name']['last']}
                Fighter 2: {fighter2['name']['first']} {fighter2['name']['last']}
                Fighter 1 bouts: {len(history1)}
                Fighter 2 bouts: {len(history2)}
                Predict this fight round by round.
                """

    # Initialize model client (DeepSeek 70B)
    client = InferenceClient(
        token=hf_token.token,
        model="deepseek-ai/DeepSeek-R1-Distill-Llama-70B"
    )

    # ---------------------------------------------------
    # Normalize chat history for LLM format
    # ---------------------------------------------------
    fixed_history = []
    for h in history:
        user_msg = h.get("user") or h.get("content") or ""
        bot_msg = h.get("assistant") or h.get("message") or ""

        if user_msg:
            fixed_history.append({"role": "user", "content": user_msg})
        if bot_msg:
            fixed_history.append({"role": "assistant", "content": bot_msg})

    # ---------------------------------------------------
    # Build entire message list
    # ---------------------------------------------------
    messages = [{"role": "system", "content": system_message}]
    messages.extend(fixed_history)
    messages.append({"role": "user", "content": message})

    # ---------------------------------------------------
    # Stream model response token-by-token
    # ---------------------------------------------------
    response_text = ""

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

        response_text += token
        yield response_text



# -------------------------------------------------------
# GRADIO CHAT UI
# -------------------------------------------------------
chatbot = gr.ChatInterface(
    respond,
    type="messages",
    additional_inputs=[
        gr.Textbox(value=system_prompt, label="System Prompt"),
        gr.Slider(1, 4096, value=800, step=1, label="Max Tokens"),
        gr.Slider(0.1, 2.0, value=0.7, step=0.1, label="Temperature"),
        gr.Slider(0.1, 1.0, value=0.95, step=0.05, label="Top-p"),
    ],
    title="🥊 BOXTRON-AI — Advanced Boxing Prediction Engine",
    description="DeepSeek-powered boxing analyst that predicts fights, simulates rounds, scores bouts, and breaks down styles."
)


# -------------------------------------------------------
# LAUNCH APP
# -------------------------------------------------------
with gr.Blocks() as demo:
    with gr.Sidebar():
        gr.Markdown("### Login Required")
        gr.LoginButton()
    chatbot.render()

if __name__ == "__main__":
    demo.launch()
from openboxing_api import find_champion_by_name, get_bouts_for_champion

fighter1 = find_champion_by_name("Isaac Cruz")
fighter2 = find_champion_by_name("Lemont Roach")

if fighter1 and fighter2:
    history1 = get_bouts_for_champion(fighter1["championId"])
    history2 = get_bouts_for_champion(fighter2["championId"])
    
    # Build your prompt for LLM
    prompt = f"""
    Fighter 1: {fighter1['name']['first']} {fighter1['name']['last']}
    Fighter 2: {fighter2['name']['first']} {fighter2['name']['last']}
    Fighter 1 bouts: {len(history1)}
    Fighter 2 bouts: {len(history2)}
    Predict this fight round by round.
    """