app.py

import streamlit as st
from unsloth import FastLanguageModel
from transformers import AutoTokenizer
import torch

@st.cache_resource
def load_model_and_tokenizer(model_name, hf_token):
    # Load the model and tokenizer
    model, tokenizer = FastLanguageModel.from_pretrained(
        model_name=model_name,
        max_seq_length=2048,
        dtype=None,
        load_in_4bit=True,
        token=hf_token,
    )
    FastLanguageModel.for_inference(model)  # Enable optimized inference
    return model, tokenizer

def generate_solution(problem, model, tokenizer):
    # Prepare the prompt using the same format as training
    prompt_template = """Below is an instruction that describes a task, paired with an input that provides further context. 
Write a response that appropriately completes the request. 
Before answering, think carefully about the question and create a step-by-step chain of thoughts to ensure a logical and accurate response.

### Instruction:
You are a math expert. Please solve the following math problem. 

### Problem:
{}

### Solution:
<think>
{{}}
</think>
{{}}"""

    prompt = prompt_template.format(problem)

    # Tokenize and prepare input
    inputs = tokenizer(
        [prompt], 
        return_tensors="pt",
        padding=True,
    ).to("cuda")

    # Generate solution
    outputs = model.generate(
        input_ids=inputs.input_ids,
        attention_mask=inputs.attention_mask,
        max_new_tokens=1200,
        temperature=0.7,
        pad_token_id=tokenizer.eos_token_id,
        use_cache=True,
    )

    # Decode and format output
    full_response = tokenizer.decode(outputs[0], skip_special_tokens=True)
    
    # Extract the generated solution part
    try:
        solution = full_response.split("### Solution:")[1].strip()
    except IndexError:
        solution = full_response  # Fallback in case formatting fails
    
    return solution

# Streamlit app
st.title("Math Problem Solver")
hf_token = st.text_input("Enter your Hugging Face token:",type="password")
model_name = "shukdevdatta123/DeepSeek-R1-Math-Solutions"

if hf_token:
    # Load model and tokenizer
    model, tokenizer = load_model_and_tokenizer(model_name, hf_token)

    # Input for custom problem
    custom_problem = st.text_input("Enter a math problem:")

    if st.button("Generate Solution"):
        if custom_problem:
            solution = generate_solution(custom_problem, model, tokenizer)
            st.write("### Generated Solution:")
            st.write(solution)
        else:
            st.error("Please enter a math problem.")
else:
    st.warning("Please enter your Hugging Face token to load the model.")