pytorch

pytorch copied to clipboard

Preserve node's stack trace during retrace

2

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #83050 AOTAutograd retraces graph module produced by torch dynamo, this PR preserves the stack trace in the original fx.Node.

SherlockNoMad

[mta] APEX style Fused Adam

6

This PR implements an APEX style FusedAdam in PyTorch. This is different from the APEX one in that this is compatible with `torch.cuda.amp.GradScaler` by setting `_step_supports_amp_scaling` to `True` and unscales...

crcrpar

[maskedtensor] first commit, core and creation

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #82841 * #82839 * #82837 * __->__ #82836

george-qi

[Quant] Vectorize scalar remainder in quantized kernel for normalization

3

## Description This PR improves performance of quantized kernel for normalize by vectorizing scalar remainder. In the current implementation [here](https://github.com/pytorch/pytorch/blob/master/aten/src/ATen/native/quantized/cpu/kernels/QuantizedOpKernels.cpp), the computation is vectorized while the scalar remainder is handled...

Xia-Weiwen

[torchgen] Fix selective build error on custom namespace

2

Summary: Currently `SelectiveBuilder` is hardcoding namespace `aten` for operators. This is not working anymore since operators started to have custom namespaces. This fixes it. Test Plan: Rely on newly added...

larryliu0820

MPS device appears much slower than CPU on M1 Mac Pro

26

### 🐛 Describe the bug Using MPS for BERT inference appears to produce about a 2x slowdown compared to the CPU. Here is code to reproduce the issue: ```python #...

mmisiewicz

Added list clearing codegen to AOTAutograd (hidden behind config.aot_clear_list

3

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #83137 * #83122 * #82874

Chillee

[maskedtensor] adding reductions

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #82841 * __->__ #82839 * #82837 * #82836

george-qi

Propose code owners from Intel

5

This PR proposes a list of CPU-related PyTorch modules that Intel is willing to own or co-own.

jgong5

Fix reshape and empty input convolution issues for channels last memory format

1

* Fixes #78611 Reshape tensors witch are channels_last will get unexpected stride. * Fixes empty input convolution issue : when input is empty e.g. shape of (0, 3, 3, 4)...

CaoE