app.py

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

# Load tokenizer
base_model = "TinyLlama/TinyLlama-1.1B-Chat-v1.0"
tokenizer = AutoTokenizer.from_pretrained(base_model)

# Load base model in empty (meta) state and move to CPU
model = AutoModelForCausalLM.from_pretrained(
    base_model,
    torch_dtype=torch.float32,
    low_cpu_mem_usage=True,
    device_map="auto"
)
model = model.to_empty(device=torch.device("cpu"))

# Load LoRA adapter and move to CPU
model = PeftModel.from_pretrained(model, "lora_adapter", device_map="cpu")
model.eval()

# Format prompt for Python tutoring
def format_prompt(instruction):
    return f"""### SYSTEM:
You are a helpful and expert Python programming tutor.
You only answer questions related to Python programming.
If the question is unrelated to Python, say:
"Sorry, I can only answer Python-related questions."

### USER:
{instruction}

### ASSISTANT:
"""

# Generate answer
def chat(instruction):
    prompt = format_prompt(instruction)
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    outputs = model.generate(
        **inputs,
        max_new_tokens=256,
        do_sample=False,
        temperature=0.0,
        top_p=1.0,
        repetition_penalty=1.1
    )
    response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    return response.split("### ASSISTANT:")[-1].strip()

# Streamlit UI
st.set_page_config(page_title="🐍 Python Tutor Chatbot")
st.title("🐍 Python Tutor Chatbot")
st.write("Ask me Python programming questions!")

user_input = st.text_area("Your question:")

if st.button("Get Answer") and user_input.strip():
    with st.spinner("Thinking..."):
        response = chat(user_input)
        st.markdown("**Answer:**")
        st.write(response)