MINI-PYTORCH / MINI-TORCH

Mini-pytorch implemented from scratch using Python

MINI-TORCH

This is my implementation of mini-pytorch from scratch.

I followed tutorial in Cornell CS5781: Machine Learning Engineering to finish this project.

Main features:

• tensor
  • tuple index
  • auto broadcast
  • reshape, view, permute
  • auto-diff and backpropagation
  • various operators and tensor functions
• nn
  • conv1d
  • conv2d
  • dropout
  • pooling
  • Linear
  • various nonlinear activations
  • module
  • parameter
• tensor ops
  • map
  • zip
  • reduce
• acceleration
  • cpu: fast ops-parallel computation
  • gpu: cuda ops

Some visualization：

1.train process

2.loss change

3.tensor graph