import os
import torch
import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer, BitsAndBytesConfig

# --- Config ---
MODEL_PATH = os.getenv("MODEL_PATH", "WhiteRabbitNeo/WhiteRabbitNeo-2.5-Qwen-2.5-Coder-7B")
LOAD_IN_4BIT = os.getenv("LOAD_IN_4BIT", "true").lower() == "true"
MAX_NEW_TOKENS = int(os.getenv("MAX_NEW_TOKENS", 2048))
DEVICE = "cuda" if torch.cuda.is_available() else "mps" if torch.backends.mps.is_available() else "cpu"

# --- Model Setup ---
quant_config = BitsAndBytesConfig(
    load_in_4bit=LOAD_IN_4BIT,
    bnb_4bit_compute_dtype=torch.bfloat16,
    bnb_4bit_use_double_quant=True,
    bnb_4bit_quant_type="nf4",
)

model = AutoModelForCausalLM.from_pretrained(
    MODEL_PATH,
    quantization_config=quant_config if LOAD_IN_4BIT else None,
    torch_dtype=torch.bfloat16 if DEVICE != "cpu" else torch.float32,
    device_map="auto" if DEVICE != "cpu" else None,
    trust_remote_code=True,
)

tokenizer = AutoTokenizer.from_pretrained(MODEL_PATH, trust_remote_code=True)

# --- Generation Function ---
def generate_code(user_prompt, temperature=0.7, top_p=0.95, max_tokens=1024, top_k=50):
    if not user_prompt.strip():
        return "⚠️ Please enter a valid prompt."

    inputs = tokenizer(user_prompt, return_tensors="pt", truncation=True)
    inputs = {k: v.to(DEVICE) for k, v in inputs.items()}

    with torch.no_grad():
        output = model.generate(
            **inputs,
            max_new_tokens=int(max_tokens),
            do_sample=True,
            temperature=float(temperature),
            top_p=float(top_p),
            top_k=int(top_k),
            pad_token_id=tokenizer.eos_token_id,
        )

    generated_text = tokenizer.decode(output[0], skip_special_tokens=True)
    new_text = generated_text[len(user_prompt):].strip()
    safe_code = new_text.replace("```", "`\u200b``")  # Prevent Markdown escape issues
    return f"```python\n{safe_code}\n```"

# --- UI ---
with gr.Blocks(title="Spec Kit Copilot") as demo:
    gr.Markdown("### 🧠 Spec Kit Copilot — AI Code Generator (Hugging Face Space Edition)")
    with gr.Row():
        with gr.Column(scale=2):
            user_input = gr.Textbox(
                label="Describe code to generate",
                lines=4,
                placeholder="E.g., Python function to parse a JSON file and pretty-print it."
            )
            with gr.Row():
                temperature = gr.Slider(0.0, 1.0, 0.7, label="Temperature")
                top_p = gr.Slider(0.0, 1.0, 0.95, label="Top-p")
            with gr.Row():
                max_tokens = gr.Slider(256, 4096, 1024, step=128, label="Max Tokens")
                top_k = gr.Slider(0, 100, 50, label="Top-k")
            generate_btn = gr.Button("🚀 Generate Code")
        with gr.Column(scale=3):
            preview = gr.Markdown("")

    generate_btn.click(
        fn=generate_code,
        inputs=[user_input, temperature, top_p, max_tokens, top_k],
        outputs=preview,
    )

if __name__ == "__main__":
    demo.launch(server_name="0.0.0.0", server_port=int(os.getenv("PORT", 7860)))