pytorch/xla

Pull Requests Commits

Allow the override of C10 visibility.

dlibenzi committed 7 years ago

8654e507

Changed the setup.py to also build the C++ tests, and package those within the wheel so that we can run end-to-end TPU tests on Kokoro.

dlibenzi committed 7 years ago

ce8f3f6f

Support build only and make faster builds.

dlibenzi committed 7 years ago

764f8439

Remove XlaHelpers::ShapeSizes

Alex Şuhan committed 7 years ago

bebb2976

Add masked_fill to ATen XLA tensor

Alex Şuhan committed 7 years ago

3ecaf88c

Fix gather override after PyTorch change

Alex Şuhan committed 7 years ago

d23df932

Properly propagate the at::ScalarType across tensor operations.

dlibenzi committed 7 years ago

2d2a1aec

Added inplace version of trigonometric operations

vishwakftw committed 7 years ago

f4882699

Set the logical type for Data from ATen constructor

Alex Şuhan committed 7 years ago

66d2b9b2

Additional minor cleanups

Alex Şuhan committed 7 years ago

3b189a93

Consolidate index implementation in one place

Alex Şuhan committed 7 years ago

74c6d981

Transfer indices to the device of base, if needed

Alex Şuhan committed 7 years ago

df4b1925

Wrap around negative indices in the index method

Alex Şuhan committed 7 years ago

d7bd06be

Handle boolean mask indices

Alex Şuhan committed 7 years ago

b074e443

Fix behavior with null indices

Alex Şuhan committed 7 years ago

2646c701

Add index to ATen XLA tensor

Alex Şuhan committed 7 years ago

94481fa6

Use ATEN ScalarType instead of XLA PrimitiveType, to avoid type inversion conflicts.

dlibenzi committed 7 years ago

bd7be5e5

Added ability to reset RNG seed.

dlibenzi committed 7 years ago

9b3c91a5

Made autogen handle cases similar to copy(), where a different "reference param" needs to be picked for device parameter.

dlibenzi committed 7 years ago

a650b631

Added inplace version of exp, expm1, log, log2, log10, log1p

vishwakftw committed 7 years ago

76e0dd26

Handle copy() operation directly.

dlibenzi committed 7 years ago

1904677d

Remove a core PyTorch patch which landed

Alex Şuhan committed 7 years ago

00c9919e

Alex Şuhan committed 7 years ago

758d1106

Remove pun-to-CPU for aten::topk not sorted.

dlibenzi committed 7 years ago

0a3846e6

Normalize positions and dimensions for certain operators.

dlibenzi committed 7 years ago

047620d5

Added aten::topk operation.

dlibenzi committed 7 years ago

a2581552

Added aten::kthvalue() operation.

dlibenzi committed 7 years ago

80722f3e

Add at::prod family of operations

vishwakftw committed 7 years ago

9978d495

Move GetCanonicalDimensionIndex to XlaHelpers

asuhan committed 7 years ago

637d0270

Add unbind to ATen XLA tensor

asuhan committed 7 years ago

9b8c33dc

Older