pytorch

pytorch copied to clipboard

[NNC] enable bf16 for mkldnn prepack conv2d

1

## Pitch Enable bf16 support for mkldnn prepack conv2d in NNC. ## Performance The BF16 conv performance has been evaluated in https://github.com/pytorch/pytorch/pull/82705. ## Additional context This PR depends on BF16...

chunyuan-w

[opinfo] conv3d

3

Reference: #74613 cc @mruberry @kshitij12345!

khushi-411

OpInfo: use functools.partial to decrease noise in make_tensor calls

3

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #84631 * __->__ #84455 * #84567 * #84454 * #84554 This `make_arg` convention removes a lot of visual clutter in the sample input...

peterbell10

Exposing native _scaled_dot_product_attention to torch.nn

15

# Summary This exposes the _scaled_dot_product_attention function to python in the nn namespace. It is still underscored because the api for args, and kwargs is still in flux for the...

drisspg

resize_as_sparse support all compressed layouts

2

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #85379 * __->__ #85378 * #85308 * #85307

amjames

Removed None arg check in test/test_decomp.py

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * #85403 * __->__ #85402 Not sure why this check was necessary? Tests seem to run fine without it. There were definitely tests this...

fdrocha

Use fallback approach for nested matmul

2

Stack from [ghstack](https://github.com/ezyang/ghstack): * __->__ #85311

mikaylagawarecki

Reference implementation for torch.Tensor.sum_to_size

1

New ref: `torch._refs.sum_to_size`. View consistency validation is disabled because the ref returns a view instead of returning the input.

IvanYashchuk

[quant][core][feature] Implement index_put for quantized CUDA tensors

1

Stack from [ghstack](https://github.com/ezyang/ghstack) (oldest at bottom): * __->__ #85164 * #85108 Summary: - Add new cuda test for quantized index_put - Add determinsitc test for CPU and CUDA quantized index_put...

jcaip

Move functorch C++ into aten/src/ATen/functorch

1

This PR moves functorch C++ code that does not depend on python into aten/src/ATen/functorch. The C++ code that does depend on python (the python bindings as well as torchdim) will...

zou3519