pytorch/xla

Pull Requests Commits

Make Cast IR op use the convert facility.

dlibenzi committed 6 years ago

6a172d9a

Adjust to Undefined removal in core PyTorch

asuhan committed 6 years ago

d10e3f4d

Fix rand() infinite recursion with XLA device

asuhan committed 6 years ago

b09c611c

New C10 visibility fix ... once again.

dlibenzi committed 6 years ago

87c6fe24

Added aten::fmod on at::Scalar type.

dlibenzi committed 6 years ago

ad4416ff

debug torchvision

Ailing Zhang committed 6 years ago

ca01ee5f

Added atan2 operation

vishwakftw committed 6 years ago

77f500c6

Add inplace version for sqrt, rsqrt, reciprocal, erf, erfinv, erfc

vishwakftw committed 6 years ago

42a3db01

Move cross implementation to tensor_ops module

asuhan committed 6 years ago

886ac817

Add cross to ATen XLA tensor

asuhan committed 6 years ago

603b5b38

Remove patch which landed

asuhan committed 6 years ago

edcf112b

Pin torchvision to the last working version

asuhan committed 6 years ago

58d5740f

Add sort to ATen XLA tensor

asuhan committed 6 years ago

a4cb8054

Allow remaining binary arithmetic operations with a non-XLA tensor

asuhan committed 6 years ago

b58cf57e

Add inplace version for sign, ceil, floor, abs and neg

vishwakftw committed 6 years ago

eff84f56

Allow the override of C10 visibility.

dlibenzi committed 6 years ago

0ff02ca0

Added aten::diag operation.

dlibenzi committed 6 years ago

645066dc

Remove unused includes from ops/select.h

asuhan committed 6 years ago

54262dd9

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 6 years ago

ce8f3f6f

Support build only and make faster builds.

dlibenzi committed 6 years ago

764f8439

Remove XlaHelpers::ShapeSizes

asuhan committed 6 years ago

bebb2976

Add masked_fill to ATen XLA tensor

asuhan committed 6 years ago

3ecaf88c

Fix gather override after PyTorch change

asuhan committed 6 years ago

d23df932

Properly propagate the at::ScalarType across tensor operations.

dlibenzi committed 6 years ago

2d2a1aec

Added inplace version of trigonometric operations

vishwakftw committed 6 years ago

f4882699

Set the logical type for Data from ATen constructor

asuhan committed 6 years ago

66d2b9b2

Additional minor cleanups

asuhan committed 6 years ago

3b189a93

Consolidate index implementation in one place

asuhan committed 6 years ago

74c6d981

Transfer indices to the device of base, if needed

asuhan committed 6 years ago

df4b1925

Wrap around negative indices in the index method

asuhan committed 6 years ago

d7bd06be

Older