Tags: wfidditch/pytorch

[nvfuser] skip extremal tests on rocm

Summary:

These are failing in rocm so disable.

[ghstack-poisoned]

Update on "SymIntify _copy functionalization kernels (and _copy_out t…

…oo)"

Signed-off-by: Edward Z. Yang <ezyangfb.com>

[ghstack-poisoned]

Update on "Generalize gesvdjBatched to run whith full_matrices==false"

As brought up in pytorch#86234 (comment), our heuristic for which SVD backend to choose was not great in some cases.
The case in which there could be some improvements is when we have a
large batch of very small non-square matrices.

This PR, adapts the calling code to gesvdj by creating two temporary
square buffers to allow to call gesvdjBatched, and then copies back the
result into the output buffers.

We then modify the heuristic that chooses between gesvdj and
gesvdjBatched.

cc VitalyFedyunin ngimel jianyuh nikitaved pearu mruberry walterddr IvanYashchuk xwang233 Lezcano JulianKnodt

Fixes pytorch#86234

[ghstack-poisoned]

Update on "[inductor] Handle nested tuple/list output in fallback ker…

…nel"

Summary: Currently fallback kernel in inductor assumes its output is
either a tensor or a tuple/list of tensors. This PR makes it handle more
generic output data structure.

[ghstack-poisoned]

update vision commit hash

Update on "Add most in-place references/decompositions"

We add most in-place references in a generic way. We also implement a
wrapper to implement the annoying interface that `nn.functional`
nonlinearities have.

We fix along the way a couple decompositions for some non-linearities by
extending the arguments that the references have.

[ghstack-poisoned]

[nn] add remove_duplicate flag to named_parameters (pytorch#88090)

Summary:
Pull Request resolved: pytorch#88090

X-link: meta-pytorch/torchrec#759

Since the remove_duplicate flag was added to named_buffers in D39493161 (pytorch@c12f829), this adds the same flag to named_parameters

Test Plan:
python test/test_nn.py -k test_buffers_and_named_buffers

OSS Tests

Reviewed By: albanD

Differential Revision: D40801899

fbshipit-source-id: 872b28f2f0bfc667ec9065353d4db45c272a8064

Update on "Enable inductor CI for TorchBench"

[ghstack-poisoned]

[Dynamo] fix torchdynamo's TVM meta schedule backend (pytorch#88249)

Note that the previous `optimize_torch` functionality of pytorch is not working with default pytorch release with  CXX11 ABI off as TVM by default needs CXX11 ABI for builds. Source: [1](https://discuss.tvm.apache.org/t/can-someone-please-give-me-the-steps-to-use-pt-tvmdsoop/12525), [2](https://discuss.pytorch.org/t/undefined-symbol-when-import-lltm-cpp-extension/32627). It would be easier for user to tune with meta schedule instead of finding a CXX11-compatible pytorch, turning on the `pt-tvmdsoop` flag in TVM and rebuilding it. This could be revisited once the `pt-tvmdsoop` flag is updated and tuned on by default in TVM.

Pull Request resolved: pytorch#88249
Approved by: https://github.com/jansel

Add support for VS2022 Windows build.