Summary:
In order to enable FC int8 quantization in P2C2, we are trying to run the caffe2 op Int8FCPackWeight in the model transformation pipeline.

The net is being generated from the python side, and passed back into C++ and run here: https://fburl.com/diffusion/3zt1mp03,  with these dependencies included: https://fburl.com/diffusion/rdjtdtcf

However, when the net is executed, it errors out with:
```
Cannot create operator of type 'Int8FCPackWeight' on the device 'CPU'
```

This diff attempts to fix this issue.

Test Plan:
To reproduce, just this test without
```
buck test //aiplatform/modelstore/transformation/tests:pyper_to_caffe2_dispatcher_test
```

Reviewed By: jspark1105

Differential Revision: D25965167

fbshipit-source-id: a7414669abb8731177c14e8792de58f400970732

Summary:
as in title

resolves D25791248 (pytorch@069602e)

Test Plan: buck test //caffe2/aten:vitals

Reviewed By: EscapeZero, malfet

Differential Revision: D26090442

fbshipit-source-id: 07937f246ec0a6eb338d21208ada61758237ae42

)

Summary:
Fixes pytorch#50378.

Additionally, this has some minor fixes:
 - [x] Fix mean for half-cauchy to return `inf` instead of `nan`.
 - [x] Fix constraints/support for the relaxed categorical distribution.

Pull Request resolved: pytorch#51053

Reviewed By: heitorschueroff

Differential Revision: D26077966

Pulled By: neerajprad

fbshipit-source-id: ca0213baa9bbdbc661aebbb901ab5e7fded38a5f

Summary: Pull Request resolved: pytorch#50884

Test Plan: Imported from OSS

Reviewed By: heitorschueroff

Differential Revision: D26086963

fbshipit-source-id: f103f7f529d63d701c4f17862e30eafbab7d0c68

Summary:
On Ampere GPU, matmuls are computed by default with TF32 when the dtype is `torch.float`:  https://pytorch.org/docs/stable/notes/cuda.html#tensorfloat-32-tf32-on-ampere-devices, which results in reduced precision in results. However, linear algebra usually need higher precision, therefore lots of tests in `test_linalg.py` are failing on Ampere GPU because of precision issue.

To fix this issue:
- Most linear algebra methods, except for matmuls, should add `NoTF32Guard`
- Expected results in unit tests should compute matmuls using numpy instead of pytorch cuda.

Pull Request resolved: pytorch#50453

Reviewed By: glaringlee

Differential Revision: D26023005

Pulled By: ngimel

fbshipit-source-id: f0ea533494fee322b07925565b57e3b0db2570c5

Summary:
Fixes #{issue number}
This is not really a new issue, just a proposed minor fix to a recent previous issue (now closed) pytorch#50640 which was a fix for pytorch#50439.

That fix added inlining for vec_signed (and others) but in one case the return was accidentally omitted.  This results in a build error:
```                 from �[01m�[K../aten/src/ATen/cpu/vec256/vec256.h:19�[m�[K,
                 from �[01m�[Katen/src/ATen/native/cpu/FillKernel.cpp.VSX.cpp:3�[m�[K:
�[01m�[K../aten/src/ATen/cpu/vec256/vsx/vsx_helpers.h:�[m�[K In function ‘�[01m�[Kvint32 vec_signed(const vfloat32&)�[m�[K’:
�[01m�[K../aten/src/ATen/cpu/vec256/vsx/vsx_helpers.h:33:1:�[m�[K �[01;31m�[Kerror: �[m�[Kno return statement in function returning non-void [�[01;31m�[K-Werror=return-type�[m�[K]
```

I've confirmed that the error disappears after this one-line fix.  (Note: There is another issue encountered later in the build unrelated to this particular fix, as I noted in a separate comment in the original issue.  I'm trying to make some sense of that one, but in any event it would be a subject for another issue/PR).

Pull Request resolved: pytorch#51116

Reviewed By: heitorschueroff

Differential Revision: D26078213

Pulled By: malfet

fbshipit-source-id: 59b2ee19138fa1b8d8ec1d35ca4a5ef0a67bc123

…ytorch#51162)

Summary:
Pull Request resolved: pytorch#51162

It's unused.
ghstack-source-id: 120427120

Test Plan: CI

Reviewed By: bhosmer

Differential Revision: D25859010

fbshipit-source-id: 7bb21312843debaedaa6a969727c171b2bb0e6b2

Summary:
Fixes pytorch#51105 by adding back the `import pycuda.autoinit`.

Pull Request resolved: pytorch#51106

Reviewed By: mingzhe09088

Differential Revision: D26086808

Pulled By: heitorschueroff

fbshipit-source-id: 88d98796c87a44cedaa1f6666e9f71a424293641

Summary:
A tiny PR to update the links in the lefthand navbar under Libraries. The canonical link for vision and text is `https://pytorch.org/vision/stable` and `https://pytorch.org/text/stable` respectively. The link without the `/stable` works via a redirect, this is cleaner.

Pull Request resolved: pytorch#51103

Reviewed By: izdeby

Differential Revision: D26079760

Pulled By: heitorschueroff

fbshipit-source-id: df1fa64d7895831f4e6242445bae02c1faa5e4dc

Summary:
Add const to static variable inside `__host__ __device__` function.

Pull Request resolved: pytorch#50970

Reviewed By: izdeby

Differential Revision: D26081478

Pulled By: heitorschueroff

fbshipit-source-id: 77cf145f7e0570359aa00aec4c8b82c950815f81

Summary:
Pull Request resolved: pytorch#51081

Pull Request resolved: pytorch#51001

fix tests in TestQuantizeJitOps

Test Plan:
Imported from OSS
python test/test_quantization.py

Reviewed By: raghuramank100

Differential Revision: D26038759

Pulled By: lyoka

fbshipit-source-id: 0977ba7b8b26a9f654f20f5c698a7a20ec078c35

Summary: Pull Request resolved: pytorch#51129

Test Plan: Imported from OSS

Reviewed By: pbelevich

Differential Revision: D26094947

Pulled By: anjali411

fbshipit-source-id: 4e1cdf8915a8c6a86ac3462685cdce881e1bcffa

…pytorch#50915)

Summary:
Pull Request resolved: pytorch#50915

Fixes pytorch#50584
Add a vectorize flag to torch.autograd.functional.jacobian and
torch.autograd.functional.hessian (default: False). Under the hood, the
vectorize flag uses vmap as the backend to compute the jacobian and
hessian, respectively, providing speedups to users.

Test Plan:
- I updated all of the jacobian and hessian tests to also use
vectorized=True
- I added some simple sanity check tests that check e.g. jacobian with
vectorized=False vs
jacobian with vectorized=True.
- The mechanism for vectorized=True goes through batched gradient
computation. We have separate tests for those (see other PRs in this
stack).

Reviewed By: heitorschueroff

Differential Revision: D26057674

Pulled By: zou3519

fbshipit-source-id: a8ae7ca0d2028ffb478abd1b377f5b49ee39e4a1

…ytorch#50356)

Summary:
Fixes pytorch#50330

- Encapsulate the `make html` call and capture the stdout/stderr with a `tee` command
- If the buld fails, print out the `WARNING:` lines of the build log and finish up with a message

I tried it out on my branch, but did not write a test.

Pull Request resolved: pytorch#50356

Reviewed By: ezyang

Differential Revision: D26101762

Pulled By: brianjo

fbshipit-source-id: ba2b704d3244ef5139ca9026c5250537bf45734f

Summary:
Fixes pytorch#49824

## Background

When creating a view of a view, there was a possibility that the new view would be less restrictive than the previous view, incorrectly sidestepping the error that should be thrown when using in-place operations on the new view.

The fix addresses this by propagating `CreationMeta` from the previous view to the new view. Currently, the old view's `creation_meta` is only propagated when the new view's `creation_meta == CreationMeta::DEFAULT`. This ensures that the new view is not less restrictive than the previous view wrt. allowing in-place operations.

Pull Request resolved: pytorch#51061

Test Plan:
```
python test/test_autograd.py TestAutogradDeviceTypeCPU.test_inplace_view_of_multiple_output_view_cpu
python test/test_autograd.py TestAutogradDeviceTypeCUDA.test_inplace_view_of_multiple_output_view_cuda
python test/test_autograd.py TestAutogradDeviceTypeCPU.test_inplace_multiple_output_view_of_view_cpu
python test/test_autograd.py TestAutogradDeviceTypeCUDA.test_inplace_multiple_output_view_of_view_cuda
```

Reviewed By: heitorschueroff

Differential Revision: D26076434

Pulled By: jbschlosser

fbshipit-source-id: c47f0ddcef9b8449427b671aff9ad08edca70fcd

Summary: Pull Request resolved: pytorch#51148

Test Plan: Imported from OSS

Reviewed By: albanD

Differential Revision: D26102025

Pulled By: anjali411

fbshipit-source-id: b1b6fd12fda03c4520a3c3200226edf352496188

Summary:
Followup of pytorch#50927

Pull Request resolved: pytorch#51045

Reviewed By: mruberry

Differential Revision: D26089204

Pulled By: ngimel

fbshipit-source-id: 77291dd83fba32d6f80a8540910b112a1d85a892

Summary:
Pull Request resolved: pytorch#51168

Adds types to function I/O for numeric suite.  This is for readability
and static type checking with mypy.

Test Plan:
```
mypy torch/quantization/
```

Imported from OSS

Reviewed By: jerryzh168

Differential Revision: D26092454

fbshipit-source-id: d37cf61e4d9604f4bc550b392f55fb59165f7624

Summary: Pull Request resolved: pytorch#50901

Test Plan: Imported from OSS

Reviewed By: ngimel

Differential Revision: D26083289

Pulled By: mruberry

fbshipit-source-id: 7e14ff37bba46dd456e0bc0aa9c4e0a632d0734c

Summary:
Mostly replace `global Foo` with `make_global(Foo)`
The only real fix is generating Subscript annotation, which is a follow up from pytorch#48676

Fixes pytorch#49617

Pull Request resolved: pytorch#51182

Reviewed By: gmagogsfm

Differential Revision: D26095244

Pulled By: malfet

fbshipit-source-id: 0e043d9a2cf43fff71dfbb341f708cd7af87c39a

Summary:
Pull Request resolved: pytorch#51079

Added support for functional conv2d + relu, will add conv1d and conv3d in future PR

Test Plan:
python test/test_quantization.py TestQuantizeFxOps.test_functional_conv

Imported from OSS

Reviewed By: vkuzo

Differential Revision: D26089964

fbshipit-source-id: 8703de17de1469f7076651c386c83fb5922a56eb

Summary:
Pull Request resolved: pytorch#50760

The SHM transport uses shared-memory-backed ringbuffers to transfer small payloads between processes on the same machine.

It was disabled in v1.6 due to a CMake mishap but we've since realized that it also doesn't work that well in docker and other setups. Enabling it here to see whether CircleCI fails.
ghstack-source-id: 120470890

Test Plan: Exported three times to CircleCI with tests consistently passing

Reviewed By: mrshenli

Differential Revision: D23814828

fbshipit-source-id: f355cb6515776debad536924de4f4d3fbb05a874

Summary:
Signed-off-by: Eli Uriegas <eliuriegas@fb.com>

Fixes #{issue number}

Pull Request resolved: pytorch#50633

Reviewed By: samestep

Differential Revision: D26083492

Pulled By: seemethere

fbshipit-source-id: c133671b9cf5074539133ee79fca5c680793a85d

…ytorch#51155)

Summary:
Pull Request resolved: pytorch#51155

This PR added support for quantizing functional conv1d, conv3d,  conv1d_relu and conv3d_relu

Test Plan:
python test/test_quantization.py TestQuantizeFxOps.test_functional_conv

Imported from OSS

Reviewed By: vkuzo

Differential Revision: D26089965

fbshipit-source-id: 4aea507d05b744807e993f6d3711ab308fb7591b

Summary:
Reference: pytorch#42515

Pull Request resolved: pytorch#50140

Reviewed By: mrshenli

Differential Revision: D25951094

Pulled By: mruberry

fbshipit-source-id: e53f1dbddff889710f05d43dbc9587382d3decb0

…ytorch#44859)

Summary:
Pull Request resolved: pytorch#44859

TensorPipe's `set_device_map` option was applied during the forward
pass. However, if we ran the backward pass for the graph we would not
automatically pick up the reverse device mapping.

As a result, users had to specify both forward and backward device mapping
which is very tedious to do.

In this PR, I've added this functionality such that TensorPipe automatically
picks up the reverse device mapping during the backward pass. This is done by
storing the appropriate device mapping in the "recv" autograd function for
distributed autograd.

#Closes: pytorch#44170
ghstack-source-id: 119950842

Test Plan:
1) waitforbuildbot
2) Unit test added.

Reviewed By: mrshenli

Differential Revision: D23751975

fbshipit-source-id: 2717d0ef5bde3db029a6172d98aad95734d52140

…onv1d and conv3d

Test Plan: revert-hammer

Differential Revision:
D26089965 (pytorch@dd1a97b)

Original commit changeset: 4aea507d05b7

fbshipit-source-id: f54184cafb9dd07858683489d8bd147474e7e4b3

Summary:
Pull Request resolved: pytorch#51180

This fast path still did a refcount bump because it copied the inner intrusive_ptr to the stack. Now it's moved.
ghstack-source-id: 120460258

Test Plan:
1) profile empty benchmark & inspect assembly to verify move
2) run framework overhead benchmarks

Reviewed By: bhosmer

Differential Revision: D26094951

fbshipit-source-id: b2e09f9ad885cb633402885ca1e61a370723f6b8

Summary:
Fixes pytorch#49542

Pull Request resolved: pytorch#50039

Reviewed By: heitorschueroff

Differential Revision: D26096247

Pulled By: ngimel

fbshipit-source-id: ec1810d3412e0d7ab6b950265a3123519ad886c1