import gradio as gr
from transformers import AutoModelForCausalLM, AutoTokenizer

# מזהה המאגר ושם קובץ ה-GGUF
MODEL_ID = "Qwen/Qwen2.5-Coder-7B-Instruct-GGUF"
GGUF_FILE = "qwen2.5-coder-7b-instruct-q4_k_m.gguf"

# 1. טעינת הטוקנייזר (ללא הפרמטר gguf_file)
tokenizer = AutoTokenizer.from_pretrained(MODEL_ID)

# 2. טעינת המודל בפורמט GGUF
model = AutoModelForCausalLM.from_pretrained(MODEL_ID, gguf_file=GGUF_FILE)

def answer(message, history):
    messages = []
    for item in history:
        if isinstance(item, dict):
            messages.append(item)
        elif isinstance(item, (list, tuple)) and len(item) == 2:
            u_msg, b_msg = item
            if u_msg: messages.append({"role": "user", "content": u_msg})
            if b_msg: messages.append({"role": "assistant", "content": b_msg})

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

    prompt = tokenizer.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)
    inputs = tokenizer(prompt, return_tensors="pt")
    
    outputs = model.generate(**inputs, max_new_tokens=512, temperature=0.7)
    response = tokenizer.decode(outputs[0][inputs.input_ids.shape[1]:], skip_special_tokens=True)
    
    return response

demo = gr.ChatInterface(
    fn=answer,
    title="Qwen GGUF Assistant",
    description="הרצת GGUF נקייה ומהירה על CPU"
)

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