"Fossies" - the Fresh Open Source Software Archive

Member "pytorch-1.8.2/GLOSSARY.md" (23 Jul 2021, 2480 Bytes) of package /linux/misc/pytorch-1.8.2.tar.gz:

As a special service "Fossies" has tried to format the requested source page into HTML format (assuming markdown format). Alternatively you can here view or download the uninterpreted source code file. A member file download can also be achieved by clicking within a package contents listing on the according byte size field.

PyTorch Glossary

Operation and Kernel


Short for "A Tensor Library". The foundational tensor and mathematical operation library on which all else is built.


A unit of work. For example, the work of matrix multiplication is an operation called aten::matmul.

Native Operation

An operation that comes natively with PyTorch ATen, for example aten::matmul.

Custom Operation

An Operation that is defined by users and is usually a Compound Operation. For example, this tutorial details how to create Custom Operations.


Implementation of a PyTorch operation, specifying what should be done when an operation executes.

Compound Operation

A Compound Operation is composed of other operations. Its kernel is usually device-agnostic. Normally it doesn't have its own derivative functions defined. Instead, AutoGrad automatically computes its derivative based on operations it uses.

Composite Operation

Same as Compound Operation.

Non-Leaf Operation

Same as Compound Operation.

Leaf Operation

An operation that's considered a basic operation, as opposed to a Compound Operation. Leaf Operation always has dispatch functions defined, usually has a derivative function defined as well.

Device Kernel

Device-specific kernel of a leaf operation.

Compound Kernel

Opposed to Device Kernels, Compound kernels are usually device-agnostic and belong to Compound Operations.

JIT Compilation


Just-In-Time Compilation.


An interface to the TorchScript JIT compiler and interpreter.


Using torch.jit.trace on a function to get an executable that can be optimized using just-in-time compilation.


Using torch.jit.script on a function to inspect source code and compile it as TorchScript code.