import torch
from transformers import AutoTokenizer, AutoModelForCausalLM
from peft import PeftModel
import streamlit as st

# Load tokenizer and base model
base_model_path = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
tokenizer = AutoTokenizer.from_pretrained(base_model_path)
base_model = AutoModelForCausalLM.from_pretrained(
    base_model_path,
    torch_dtype=torch.float16 if torch.cuda.is_available() else torch.float32,
    device_map="auto" if torch.cuda.is_available() else None
)

# Load LoRA Adapter
model = PeftModel.from_pretrained(base_model, "lora_adapter")
model.eval()

# Streamlit UI
st.title("🧠 TinyLLaMA Python Tutor (LoRA)")
st.write("Ask me any **Python programming** question:")

user_input = st.text_input("Your question")

if user_input:
    # Better Prompt Template
    prompt = f"""You are a helpful Python programming tutor.

You will ONLY answer questions related to Python programming.
If the question is unrelated to Python, reply:
"Sorry, I can only answer Python-related questions."

Question: {user_input}
Answer:"""

    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)

    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=300,
            temperature=0.7,
            do_sample=True,
            top_p=0.95,
            eos_token_id=tokenizer.eos_token_id,
            pad_token_id=tokenizer.eos_token_id
        )

    decoded_output = tokenizer.decode(outputs[0], skip_special_tokens=True)

    # Extract answer after 'Answer:' line
    answer_start = decoded_output.find("Answer:")
    if answer_start != -1:
        final_answer = decoded_output[answer_start + len("Answer:"):].strip()
    else:
        final_answer = decoded_output.strip()

    st.markdown(f"**Answer:** {final_answer}")