Jeffrey Wan

Fix and unskip dataloader tests on ARM

1

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

Fix and unskip cpp extension tests for ARM

1

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

Set default qengine to QNNPACK on ARM for quantization tests

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #83125 * #83115 * __->__ #83097 * #83089 The issue is that there is no FBGEMM or MKLDNN on ARM and QNNPACK is...

Unskip CompositeCompliance tests for ARM

1

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

`torch.autograd.gradcheck.get_numerical_jacobian` in `utils.gradcheck` will be deprecated in PT 1.9

2

Update: we've actually decided to go through the usual deprecation process for this, so it won't be removed immediately Due to changes in the upcoming 1.9 release, https://github.com/NVIDIA/MinkowskiEngine/blob/master/MinkowskiEngine/utils/gradcheck.py#L35 which relies...

[Reland] aot autograd explicitly errors on double backward

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #92893 This reverts commit fb980581a7b41a5ea570fcb03829463b806b3bbc. Testing: `python benchmarks/dynamo/timm_models.py --float32 --training --only=mobilevit_s --performance --inductor --disable-cudagraphs` ``` main: memory: eager: 12.30 GB, dynamo: 12.28...

cc @sanchitintel @jingxu10

[do not review] saving stuff

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #125326 * __->__ #125270 cc @ezyang @msaroufim @bdhirsh @anijain2305 @chauhang @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng

[do not review] Saving more things

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #125326 * #125270 cc @voznesenskym @penguinwu @EikanWang @jgong5 @Guobing-Chen @XiaobingSuper @zhuhaozhe @blzheng @wenzhe-nrv @jiayisunx @chenyang78 @kadeng @chauhang

Support custom layout call under torch dispatch mode

5

Fixes #ISSUE_NUMBER