app.py

# streamlit_app.py
import streamlit as st
import re
from sympy import symbols, integrate, exp, pi
from transformers import AutoTokenizer, AutoModelForCausalLM
import torch

st.set_page_config(page_title="AI Physics Solver", page_icon="🧠")

x, t = symbols("x t")

def extract_integral(problem_text):
    match = re.search(r'(\d+)\*?[tx]\^(\d+)', problem_text)
    limits = re.findall(r'[tx]\s*=\s*([\d\.\w]+)', problem_text)
    exp_match = re.search(r'(\d+)e\^([\-+]?[\d\.]*)[tx]', problem_text)

    if 'radioactive' in problem_text or 'half-life' in problem_text:
        decay_match = re.search(r'(\d+)\s*e\^\s*-\s*(\d+\.?\d*)[tx]', problem_text)
        if decay_match and len(limits) == 2:
            N0 = int(decay_match.group(1))
            lam = float(decay_match.group(2))
            lower, upper = map(lambda v: eval(v, {"pi": pi}), limits)
            expr = lam * N0 * exp(-lam * t)
            return f"Total decayed = {integrate(expr, (t, lower, upper)).evalf()} units."

    if match and len(limits) == 2:
        coefficient = int(match.group(1))
        exponent = int(match.group(2))
        lower_limit = eval(limits[0], {"pi": pi})
        upper_limit = eval(limits[1], {"pi": pi})
        expr = coefficient * x**exponent
        return f"Accumulated Quantity = {integrate(expr, (x, lower_limit, upper_limit))}"

    return "Could not parse the integral format."

@st.cache_resource
def load_deepseek():
    model_name = "deepseek-ai/deepseek-math-7b-base"
    tokenizer = AutoTokenizer.from_pretrained(model_name)
    model = AutoModelForCausalLM.from_pretrained(
        model_name,
        torch_dtype=torch.float16,
        device_map="auto"
    )
    return tokenizer, model

def run_deepseek(user_question):
    tokenizer, model = load_deepseek()

    solution_steps = """### Solution Format:
1. **Identify Given Data**
2. **Determine the Required Variable**
3. **Apply Relevant Equations**
4. **Substitute Values & Solve Step-by-Step**
5. **Verify & Conclude**

### Final Answer Format:
Final Answer: [VARIABLE] = [ANSWER] [UNIT]
"""
    prompt = f"Q: Solve the following physics problem using rigorous mathematical reasoning. Do not skip any steps.\n\nProblem: {user_question}\n\n{solution_steps}\nA:"
    inputs = tokenizer(prompt, return_tensors="pt").to("cuda")

    with torch.no_grad():
        outputs = model.generate(
            **inputs,
            max_new_tokens=500,
            temperature=0.1,
            repetition_penalty=1.0,
            eos_token_id=tokenizer.eos_token_id,
            pad_token_id=tokenizer.eos_token_id
        )
    return tokenizer.decode(outputs[0], skip_special_tokens=True).split("A:")[-1].strip()

# ---------------- UI Layout ----------------
st.title("🧠 AI Science Solver")

task_type = st.selectbox("Choose Task Type", ["LLM Reasoning (DeepSeek)", "Symbolic Integration"])
user_question = st.text_area("Enter your physics or math question below:")

if st.button("Solve"):
    with st.spinner("Solving..."):
        if task_type == "LLM Reasoning (DeepSeek)":
            result = run_deepseek(user_question)
        else:
            result = extract_integral(user_question)

    st.subheader("📘 Answer")
    st.write(result)