pytorch

pytorch copied to clipboard

[Profiler] Start moving python bindings out of autograd

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #83033 * #82993 * #82992 * __->__ #82584 A lot of profiler code still lives in autograd for historic reasons. However as we...

robieta

Fix and unskip cpp extension tests for ARM

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #83125 * __->__ #83115 * #83097 * #83089

soulitzer

enable channels last 3d for Conv3d and ConvTranspose3d on mkldnn path

5

Stack from [ghstack](https://github.com/ezyang/ghstack): * __->__ #74023 * #70897 * #77060 Differential Revision: [D35782442](https://our.internmc.facebook.com/intern/diff/D35782442) Add channels last 3d support for both Conv3d and ConvTransposed3d

mingfeima

add channels last support for slow_conv_transpose2d

10

Stack from [ghstack](https://github.com/ezyang/ghstack): * #74023 * __->__ #70897 * #77060 Differential Revision: [D33571076](https://our.internmc.facebook.com/intern/diff/D33571076) This patch is about enabling `channels last` support for the fallback ATen implementation for transposed convolution. So...

mingfeima

opitimze ConvTransposedND with mkldnn float32 and bfloat16 on CPU

6

Stack from [ghstack](https://github.com/ezyang/ghstack): * #74023 * #70897 * __->__ #77060 Differential Revision: [D36377117](https://our.internmc.facebook.com/intern/diff/D36377117) This patch is about enabling mkldnn on float32 and bfloat16 for transposed convolution on CPU path.

mingfeima

[RFC] TorchStore - A Shared-Memory Tensor Store

36

## 🚀 Feature TorchStore is a key-value store that holds ATen tensors in shared memory so that they can be accessed across process boundaries without any expensive copy operations. ##...

cbalioglu

### 🐛 Describe the bug https://github.com/pytorch/pytorch/blob/b2363520363d8c85496f6954fde005834c188b5d/torch/nn/modules/transformer.py#L358-L359 https://github.com/pytorch/pytorch/blob/b2363520363d8c85496f6954fde005834c188b5d/torch/nn/modules/transformer.py#L563-L565 ### Versions We don't need a version to fix a typo.

developer0hye

Automated submodule update: kineto

7

This is an automated pull request to update the first-party submodule for [pytorch/kineto](https://github.com/pytorch/kineto). New submodule commit: https://github.com/pytorch/kineto/commit/066696ef46f17709606d02a824658a1b4430e3ad Test Plan: Ensure that CI jobs succeed on GitHub before landing.

facebook-github-bot

[Profiler] Break metadata generation into multiple visitors

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #83033 * #82993 * #82992 * #82584 This is a no-op change which establishes a base class to handle Result to Kineto...

robieta

Update ideep for NNC post-op

4

### Description This PR is to add NNC post-op fusion support in ideep for further NNC development. It includes: - element wise post op fusion - conv/matmal/linear + binary post...

yanbing-j