import gradio as gr
import spaces
import re
from huggingface_hub import hf_hub_download
from llama_cpp import Llama


MODEL_REPO = "HauhauCS/Qwen3.5-9B-Uncensored-HauhauCS-Aggressive"
MODEL_FILE = "Qwen3.5-9B-Uncensored-HauhauCS-Aggressive-Q4_K_M.gguf"

SYSTEM_PROMPT = """You are Eagle Vision, a highly sophisticated AI intelligence entity model.
Answer naturally, clearly, accurately, and completely.
Respond in the same language used by the user.
Maintain conversational continuity and do not unnecessarily shorten answers."""


llm = None


def load_model():
    global llm

    if llm is None:
        model_path = hf_hub_download(
            repo_id=MODEL_REPO,
            filename=MODEL_FILE
        )

        llm = Llama(
            model_path=model_path,
            n_ctx=8192,
            n_gpu_layers=-1,
            verbose=False
        )

    return llm


@spaces.GPU(duration=120)
def eagle_vision(message, history):
    model = load_model()

    messages = [
        {
            "role": "system",
            "content": SYSTEM_PROMPT
        }
    ]

    for user_message, assistant_message in history:
        messages.append({
            "role": "user",
            "content": user_message
        })

        messages.append({
            "role": "assistant",
            "content": assistant_message
        })

    messages.append({
        "role": "user",
        "content": message
    })

    response = model.create_chat_completion(
        messages=messages,
        max_tokens=2048,
        temperature=0.7,
        top_p=0.8,
        top_k=20
    )

    result = response["choices"][0]["message"]

    content = result.get("content", "")
    reasoning = result.get("reasoning_content", "")

    if not content and reasoning:
        content = reasoning

    content = re.sub(
        r"^\s*Thinking Process:.*?</think>\s*",
        "",
        content,
        flags=re.DOTALL | re.IGNORECASE
    )

    content = re.sub(
        r"<think>.*?</think>",
        "",
        content,
        flags=re.DOTALL | re.IGNORECASE
    )

    content = re.sub(
        r"</think>",
        "",
        content,
        flags=re.IGNORECASE
    )

    return content.strip()


demo = gr.ChatInterface(
    fn=eagle_vision,
    title="EAGLE VISION",
    description="Universal Intelligence"
)


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