README.md

---
base_model: unsloth/meta-llama-3.1-8b-instruct-bnb-4bit
language:
- en
license: mit
tags:
- text-generation-inference
- transformers
- unsloth
- llama
- trl
datasets:
- TIGER-Lab/MathInstruct
library_name: transformers
---

# Model Card: Math Homework Solver

This is a Large Language Model (LLM) fine-tuned to solve math problems with detailed, step-by-step explanations and accurate answers. The base model used is Llama 3.1 with 8 billion parameters, which has been quantized to 4-bit using QLoRA (Quantized Low-Rank Adaptation) and PEFT (Parameter-Efficient Fine-Tuning) through the Unsloth framework.

Other Homework Solver Models include [Science_Homework_Solver_Llama318B](https://huggingface.co/justsomerandomdude264/Science_Homework_Solver_Llama318B) and [SocialScience_Homework_Solver_Llama318B](https://huggingface.co/justsomerandomdude264/SocialScience_Homework_Solver_Llama318B)

## Model Details

- **Base Model**: Llama 3.1 (8 Billion parameters)
- **Fine-tuning Method**: PEFT (Parameter-Efficient Fine-Tuning) with QLoRA
- **Quantization**: 4-bit quantization for reduced memory usage
- **Training Framework**: Unsloth, optimized for efficient fine-tuning of large language models
- **Training Environment**: Google Colab (free tier), NVIDIA T4 GPU (16GB VRAM), 12GB RAM
- **Dataset Used**: TIGER-Lab/MathInstruct (Yue, X., Qu, X., Zhang, G., Fu, Y., Huang, W., Sun, H., Su, Y., & Chen, W. (2023). MAmmoTH: Building Math Generalist Models through Hybrid Instruction Tuning. *arXiv preprint arXiv:2309.05653*.
), 560 selected math problems and solutions
- **Git Repo**: The git repo on my github account is [justsomerandomdude264/Homework_Solver_LLM](https://github.com/justsomerandomdude264/Homework_Solver_LLM)

## Capabilities

The Math Homework Solver model is designed to assist with a broad spectrum of mathematical problems, from basic arithmetic to advanced calculus. It provides clear and detailed explanations, making it an excellent resource for students, educators, and anyone looking to deepen their understanding of mathematical concepts.

By leveraging the Llama 3.1 base model and fine-tuning it using PEFT and QLoRA, this model achieves high-quality performance while maintaining a relatively small computational footprint, making it accessible even on limited hardware.

## Getting Started

To start using the Math Homework Solver model, follow these steps:

1. **Clone the repo**
   ```bash
   git clone https://huggingface.co/justsomerandomdude264/Math_Homework_Solver-Llama3.18B
   ```
   
2. **Run inference**

   1. This method is recommended as its reliable and accurate:
   ```python
   from unsloth import FastLanguageModel
   import torch
    
   # Define Your Question
   question = "Verify that the function y = a cos x + b sin x, where, a, b ∈ R is a solution of the differential equation d2y/dx2 + y=0." # Example Question, You can change it with one of your own

   # Load the model
   model, tokenizer = FastLanguageModel.from_pretrained(
           model_name = "Math_Homework_Solver_Llama318B/model_adapters", # The dir where the repo is cloned or "\\" for root
           max_seq_length = 2048,
           dtype = None,
           load_in_4bit = True,
       )

   # Set the model in inference model
   FastLanguageModel.for_inference(model)

   # QA template
   qa_template = """Question: {}
   Answer: {}"""

   # Tokenize inputs
   inputs = tokenizer(
   [
       qa_template.format(
           question, # Question
           "", # Answer - left blank for generation
       )
   ], return_tensors = "pt").to("cuda")

   # Stream the answer/output of the model
   from transformers import TextStreamer
   text_streamer = TextStreamer(tokenizer)
   _ = model.generate(**inputs, streamer = text_streamer, max_new_tokens = 512)
   ```
   
   2. Another way to run inference is to use the merged adapters (not recommend as it gives inaccurate/different answers):
   ```python
   from transformers import LlamaForCausalLM, AutoTokenizer

   # Load the model
   model = LlamaForCausalLM.from_pretrained(
       "justsomerandomdude264/Math_Homework_Solver_Llama318B",
       device_map="auto"
   )
   
   # Load the tokenizer
   tokenizer = AutoTokenizer.from_pretrained("justsomerandomdude264/Math_Homework_Solver_Llama318B")

   # Set the inputs up
   qa_template = """Question: {}
   Answer: {}"""
   
   inputs = tokenizer(
   [
       qa_template.format(
           "Verify that the function y = a cos x + b sin x, where, a, b ∈ R is a solution of the differential equation d2y/dx2 + y=0.", # Question
           "", # output - leave this blank for generation!
       )
   ], return_tensors = "pt").to("cuda")
   
   # Do a forward pass
   outputs = model.generate(**inputs, max_new_tokens = 128, use_cache = True)
   raw_output = str(tokenizer.batch_decode(outputs))
   
   # Formtting the string
   # Removing the list brackets and splitting the string by newline characters
   formatted_string = raw_output.strip("[]").replace("<|begin_of_text|>", "").replace("<|eot_id|>", "").strip("''").split("\\n")

   # Print the lines one by one
   for line in formatted_string:
     print(line)
   ```
  

## Citation

Please use the following citation if you reference the Math Homework Solver model:

### BibTeX Citation
```bibtex
@misc{paliwal2024,
    author = {Krishna Paliwal},
    title = {Contributions to Math_Homework_Solver},
    year = {2024},
    email = {krishna.plwl264@gmail.com}
}
```

### APA Citation
```plaintext
Paliwal, Krishna (2024). Contributions to Math_Homework_Solver. Email: krishna.plwl264@gmail.com .
```