MiniTorch

Chenran (cn257)

This project is Chenran's implemented version of the course project minitorch at Cornell Tech.

Instructed by Professor Sasha Rush

Project Descriptions

• Implemented a deep learning training tool from scratch using Python, designed efficient data structures for tensor operations and related deep learning functionalities, to enable users to build and train deep learning models.

• Developed fundamental modules such as auto-differentiation and backpropagation, broadcasting, GPUs and Parallel Programming,

• Optimized data structures to improve computational efficiency and memory usage, and created multiple test cases to validate the correctness of the functions.

Project Overview

from minitorch page

• Docs: https://minitorch.github.io/

• Overview: https://minitorch.github.io/module4.html

This module requires fast_ops.py, cuda_ops.py, scalar.py, tensor_functions.py, tensor_data.py, tensor_ops.py, operators.py, module.py, and autodiff.py from Module 3.

Additionally you will need to install and download the MNist library.

(On Mac, this may require installing the wget command)

pip install python-mnist
mnist_get_data.sh

• Tests:

python run_tests.py

This assignment requires the following files from the previous assignments. You can get these by running

python sync_previous_module.py previous-module-dir current-module-dir

The files that will be synced are:

    minitorch/tensor_data.py minitorch/tensor_functions.py minitorch/tensor_ops.py minitorch/fast_ops.py minitorch/cuda_ops.py minitorch/operators.py minitorch/module.py minitorch/autodiff.py minitorch/module.py project/run_manual.py project/run_scalar.py project/run_tensor.py project/run_fast_tensor.py project/parallel_check.py tests/test_tensor_general.py