Summary:
`test_neg` sometimes fails internally because `random_()` can generate an out-of-range value for CharTensor. This PR fixes it.
Pull Request resolved: pytorch#9431

Reviewed By: SsnL

Differential Revision: D8843284

Pulled By: yf225

fbshipit-source-id: bf516cceb8f780e133fa54f7364c77821eb7c013

Summary:
If the type_as operator takes in two values with the same type, remove that operator.
Pull Request resolved: pytorch#9316

Reviewed By: zdevito

Differential Revision: D8808355

fbshipit-source-id: 2d5710a6380b22f4568fc38a439061b5340c4eb1

Reviewed By: apaszke

Differential Revision: D8863572

Pulled By: ezyang

fbshipit-source-id: 13749f51320a4e403644674b0335aed4987fa887

Summary:
The tests were too flaky, and the procedure for legitimately
updating versions of software too onerous, to warrant continually
testing these.

Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Pull Request resolved: pytorch#9459

Reviewed By: zou3519

Differential Revision: D8852357

Pulled By: ezyang

fbshipit-source-id: 24e99cd00b4252cdeec2a1d9af92456b4a54912a

Summary:
Add `WeakTensor` - a `Tensor` counterpart which doesn't keep the data (or any other expensive resources) alive. They can be `.lock()`ed and return `at::optional<Tensor>` if they're still alive.
Pull Request resolved: pytorch#9363

Reviewed By: ezyang

Differential Revision: D8815434

Pulled By: apaszke

fbshipit-source-id: 1b3e96503c1285d78ef124c585e65c7630f3253e

…orch#9368)

Summary:
IValue is short for interpreter value. It is used frequently so a short name is important.
This will allow us to implement more non-tensor types in an efficient way and remove
many hacks from the compiler.

This PR is limited. It only introduces IValue and changes interpreter to use it.
Follow up PRs will:
* Change the way aten_ops consume non-tensor types so that integer lists,
  are no longer represented as Tensors.
* Introduce TensorList as a fundamental type and remove all vararg handling in gen_jit_dispatch
* Change the compiler to implement math on primitive numbers rather than converting to tensors.

jamesr66a  apaszke
Pull Request resolved: pytorch#9368

Reviewed By: ezyang

Differential Revision: D8817598

Pulled By: zdevito

fbshipit-source-id: 29dce80611ce5f6384234de9d12a67861d2b112f

Summary:
Storage views were previously used to implement CUDA IPC sharing,
but they weren't necessary.  The new strategy is described in
Note [CUDA IPC and the caching allocator].

This also fixes an unrelated bug, where we weren't actually using
the Tensor forking pickler, because we didn't register a pickler
for torch.Tensor.

Fixes pytorch#9447.  Fixes ROCm#46.

Signed-off-by: Edward Z. Yang <ezyang@fb.com>

CC apaszke
Pull Request resolved: pytorch#9466

Reviewed By: apaszke

Differential Revision: D8859698

Pulled By: ezyang

fbshipit-source-id: 3362cb92f6ae4aa37084c57d79b31004bd0b4a97

Summary:
Signed-off-by: Edward Z. Yang <ezyang@fb.com>
Pull Request resolved: pytorch#9467

Reviewed By: houseroad

Differential Revision: D8860794

Pulled By: ezyang

fbshipit-source-id: 9b11475d9bb4b3361973865d7f68e562bffbf9d8

Summary:
fixes pytorch#9409
Pull Request resolved: pytorch#9425

Reviewed By: ezyang

Differential Revision: D8842844

Pulled By: wanchaol

fbshipit-source-id: 3c6b26470d59d8d1fc5f79caa70252b9de7290e4

Summary:
Update onnx-tensort to follow up recent changes.
Pull Request resolved: pytorch#9469

Reviewed By: Maratyszcza

Differential Revision: D8866704

Pulled By: yinghai

fbshipit-source-id: 3b96ec2fa28470f0d4b5a7c62ab332eeba4bdb12

Summary: Pull Request resolved: pytorch#9473

Reviewed By: houseroad

Differential Revision: D8865754

Pulled By: bddppq

fbshipit-source-id: 406eda6c145f03a0ee35c4643ec1ec0092fbce88

Summary:
Pull Request resolved: pytorch#9153

Closes pytorch#9153

Modified the values reported by the benchmarking platform to include tensor_shape and op_args. These values have a different naming scheme to values like flops and latency.

Reviewed By: sf-wind

Differential Revision: D8729791

fbshipit-source-id: f050200be01c6d0794bf5faaa6e8cef12a00affe

Summary:
This fixes pytorch#9054.
Pull Request resolved: pytorch#9398

Reviewed By: ezyang

Differential Revision: D8827778

Pulled By: yf225

fbshipit-source-id: 862287802cb69c6ac71ff4df19cadb89b1face1d

…compatibility (pytorch#9403)

Summary:
Pull Request resolved: pytorch#9403

In BBoxTransform and GenerateProposal ops, clip_boxes makes sure the bbox fits
within the images. For rotated boxes, this doesn't always make sense as there
could be multiple ways to clip a rotated box within an image boundary.
Moreover, clipping to a horizontal box means we leave out pixels of interest
potentially. Therefore, we clip only boxes with angle almost equal to 0 (with a
specified `angle_thresh` tolerance).

Reviewed By: pjh5

Differential Revision: D8828588

fbshipit-source-id: 39c1eafdb5d39d383780faa0a47e76149145e50c

Summary:
Enable fusion for IDEEP in optimizeForIdeep
including Conv+ReLU, Conv+Sum, Conv+Sum+ReLU, Conv+BN
Pull Request resolved: pytorch#9255

Reviewed By: bddppq

Differential Revision: D8809030

Pulled By: yinghai

fbshipit-source-id: af30bad3b96cb965bd26a4dfa810370faec4bb88

Summary:
I noticed that `Sequential::clone()` does not work. This is because `Sequential` does not use `reset()` which is normally where modules have to initialize and register its submodules. Further, this is because of the way `Sequential` allows its modules to be passed in the constructor, which doesn't work with `reset()` (since it does "late" initialization).

I've added some better error messages inside `Cloneable::clone()` which makes this kind of mistake clearer for other users, and tests for `Sequential::clone()`.

I also had to give `AnyModule` a deep `clone()` method.

ebetica ezyang
Pull Request resolved: pytorch#9372

Differential Revision: D8865189

Pulled By: goldsborough

fbshipit-source-id: b81586e0d3157cd3c4265b19ac8dd87c5d8dcf94

Summary:
onnx/onnx@b2817a6
Pull Request resolved: pytorch#9476

Reviewed By: houseroad

Differential Revision: D8868253

Pulled By: bddppq

fbshipit-source-id: b1f14bab47f020f0bc0239da7e2bbf959a407d6a

Summary:
This change makes README.md compatible with both Github and VSTS markdown engines. Images can be reduced if necessary
Pull Request resolved: pytorch#9296

Differential Revision: D8874931

Pulled By: soumith

fbshipit-source-id: 0c530c1e00b06fc891301644c92c33007060bf27