pytorch/xla

Pull Requests Commits

Bump up the IR graph trim thresholds.

dlibenzi committed 7 years ago

a36c90c3

Extend the graph-saving debug utility to more places where we enter the XLA tensor world for execution.

dlibenzi committed 7 years ago

25d099bf

Read async status in wait operation.

dlibenzi committed 7 years ago

f3221842

Remove Hash() default to DataHash(), and restrict to arithmetic for it.

dlibenzi committed 7 years ago

b2f6d951

Fix code by coping with the fact that ATEN copy() is no more.

dlibenzi committed 7 years ago

5dd0ea12

Return a torch.device consistently.

dlibenzi committed 7 years ago

d2e239b0

Fix index_put starting with null indices

Alex Şuhan committed 7 years ago

64de8e6a

Add base support for ATEN Bool tensor type.

dlibenzi committed 7 years ago

04faa305

Send tensors to XLA device before running tests, in case of tensor-modifying test functions.

dlibenzi committed 7 years ago

9199f36f

Made aten::clamp operation to use possibly cache min/max arguments, w/out generating ops::Scalar.

dlibenzi committed 7 years ago

82e82f5a

Fix multi-index operation which starts with null indices

Alex Şuhan committed 7 years ago

0c543068

Make unsqueeze() return a view.

dlibenzi committed 7 years ago

5b6085ce

Promote sides of bitwise logical operations

Alex Şuhan committed 7 years ago

efd20643

Add randperm to ATen XLA tensor

Alex Şuhan committed 7 years ago

002eb2d3

Fixed zero-sized slice operation.

dlibenzi committed 7 years ago

695c93f7

Fix unsqueeze operation symbol.

dlibenzi committed 7 years ago

1d95c624

Add s__th_eq wrapper to ATen XLA tensor

Alex Şuhan committed 7 years ago

04fd6ddd

Route more scalars to (possibly) device data.

dlibenzi committed 7 years ago

3086d260

Add CircleCI badge to README file.

dlibenzi committed 7 years ago

0e359885

Fix README file WRT extra DEB repo addition.

dlibenzi committed 7 years ago

ebafd8c9

Revert manually changes with patch-R.

dlibenzi committed 7 years ago

48e34864

Remove TestGradients due to breakage on the PT side, and due to close proximity to JIT-abandoning point.

dlibenzi committed 7 years ago

1f4006bb

Add tests for index with integer base

Alex Şuhan committed 7 years ago

c58c087c

Add more shape checks

Alex Şuhan committed 7 years ago

14069f15

Fix s_copy_ and indexing types

Alex Şuhan committed 7 years ago

16cafcff

Add GraphExecutor bindings removed from torch to torch_xla

asuhan committed 7 years ago

c46e02df

Promote binary operations which are going to be numeric.

dlibenzi committed 7 years ago

36403e8d

Add Debian Testing source to README.

dlibenzi committed 7 years ago

fea07d23

Do not rely only on the optimizer state to sync the tensors.

dlibenzi committed 7 years ago

bfe012ee

Use tf2xla helpers to implement convolution gradients

asuhan committed 7 years ago

2fcdb6cd

Older