PT2 extension points

Update: 2024-02-05

Description

We discuss some extension points for customizing PT2 behavior across Dynamo, AOTAutograd and Inductor.

Comments

In Channel

Compiler collectives

2024-08-0416:33

TORCH_TRACE and tlparse

2024-04-2915:28

Higher order operators

2024-04-2117:10

Inductor - Post-grad FX passes

2024-04-1224:07

CUDA graph trees

2024-03-2420:50

Min-cut partitioner

2024-03-1715:56

AOTInductor

2024-03-0217:30

Tensor subclasses and PT2

2024-02-2413:25

Compiled autograd

2024-02-1918:07

PT2 extension points

2024-02-0515:54

Inductor - Define-by-run IR

2024-01-2412:06

Unsigned integers

2024-01-1713:07

Inductor - IR

2024-01-1618:00

Dynamo - VariableTracker

2024-01-1215:55

Unbacked SymInts

2023-02-2121:31

Zero-one specialization

2023-02-2021:07

torchdynamo

2022-12-0625:35

PyTorch 2.0

2022-12-0417:51

History of functorch

2022-11-0719:10

Learning rate schedulers

2022-06-1319:35

00:00

#box-pro-ellipsis-175799208158291{-webkit-line-clamp:2;}PT2 extension points

PT2 extension points

PyTorch

PT2 extension points