randtext.py

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

st.set_page_config(
    page_title="Section 8.1 Legal Assistant",
    page_icon="⚖️",
    layout="centered",
)

st.title("⚖️ Section 8.1 Legal Assistant")
st.markdown("**Reinforcement Fine-Tuned Model for ITAA 1997 - Section 8.1 (General Deductions)**")
st.markdown("---")

SYSTEM_PROMPT = """You ONLY answer questions about Section 8.1 of the Income Tax Assessment Act 1997 (General Deductions). If a question is about any other section, topic, or contains wrong details about Section 8.1, refuse or correct it. Never add information not in Section 8.1."""

MODEL_ID = "muhammadjasim12/rainforcejasim"


@st.cache_resource
def load_model():
    """Load model once and cache it."""
    tokenizer = AutoTokenizer.from_pretrained(MODEL_ID, trust_remote_code=True)
    base_model = AutoModelForCausalLM.from_pretrained(
        "Qwen/Qwen2.5-7B-Instruct",
        torch_dtype=torch.float16,
        device_map="auto",
        trust_remote_code=True,
    )
    model = PeftModel.from_pretrained(base_model, MODEL_ID)
    model.eval()
    return model, tokenizer


def ask(question, model, tokenizer):
    prompt = (
        f"<|im_start|>system\n{SYSTEM_PROMPT}<|im_end|>\n"
        f"<|im_start|>user\n{question}<|im_end|>\n"
        f"<|im_start|>assistant\n"
    )
    inputs = tokenizer(prompt, return_tensors="pt").to(model.device)
    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=300,
            do_sample=False,
            pad_token_id=tokenizer.eos_token_id,
        )
    return tokenizer.decode(
        outputs[0][inputs["input_ids"].shape[1]:],
        skip_special_tokens=True
    ).strip()


# Load model
with st.spinner("Loading model... (first time takes 2-3 minutes)"):
    model, tokenizer = load_model()

st.success("Model loaded!")
st.markdown("---")

# Chat interface
if "messages" not in st.session_state:
    st.session_state.messages = []

# Display chat history
for msg in st.session_state.messages:
    with st.chat_message(msg["role"]):
        st.markdown(msg["content"])

# User input
user_input = st.chat_input("Ask a question about Section 8.1...")

if user_input:
    # Show user message
    st.session_state.messages.append({"role": "user", "content": user_input})
    with st.chat_message("user"):
        st.markdown(user_input)

    # Get model answer
    with st.chat_message("assistant"):
        with st.spinner("Thinking..."):
            answer = ask(user_input, model, tokenizer)
        st.markdown(answer)

    st.session_state.messages.append({"role": "assistant", "content": answer})

# Sidebar
with st.sidebar:
    st.header("About")
    st.markdown("""
    This model is **reinforcement fine-tuned (DPO + SFT)** 
    exclusively on **Section 8.1** of the Income Tax 
    Assessment Act 1997 (General Deductions).
    
    **It will:**
    - Answer questions about Section 8.1
    - Refuse questions about other sections
    - Correct wrong details in questions
    
    **It will NOT:**
    - Answer questions outside Section 8.1
    - Add information not in the section
    - Make up dollar amounts or rules
    """)

    st.markdown("---")
    st.markdown("**Model:** `muhammadjasim12/rainforcejasim`")
    st.markdown("**Base:** Qwen2.5-7B-Instruct")
    st.markdown("**Training:** DPO + SFT")

    st.markdown("---")
    st.header("Example Questions")
    st.markdown("""
    - What is Section 8.1 about?
    - What does Section 8.1(1)(a) say?
    - Can I deduct a capital expense?
    - What does Section 8.2 say?
    - Does Section 8.1 have four subsections?
    - What is Division 7A?
    """)

    if st.button("Clear Chat"):
        st.session_state.messages = []
        st.rerun()