forked from pytorch/pytorch
-
Notifications
You must be signed in to change notification settings - Fork 1
PyTorch Autograd Engine
Sameer Deshmukh edited this page Nov 2, 2019
·
3 revisions
This article summarizes the architecture of AutoGrad Engine of PyTorch.
-
torch.Tensor
-
torch.autograd.Function
-
torch.autograd.gradcheck
-
torch.autograd.gradgradcheck
-
torch.autograd.profiler
A neural network or a graph has multiple nodes. Each node is a tensor. A tensor/node is called a leaf if it has enable_grad set to false.
A lot of the backward differentiation functions are generated automatically via the gen_autograd_functions.py file.