pytorch/xla

Pull Requests Commits

Added aten::any and aten::all operations.

dlibenzi committed 7 years ago

a208b844

Add reshape to ATen XLA tensor

Alex Şuhan committed 7 years ago

77656686

Complete support for einsum via PyTorch lowering

Alex Şuhan committed 7 years ago

c54fb2a8

Added aten::index_select operation.

dlibenzi committed 7 years ago

6ca69874

Add bmm to ATen XLA tensor

Alex Şuhan committed 7 years ago

782e0673

Add aten::full family and simplify code.

dlibenzi committed 7 years ago

fe0c7004

Change int in XLATensor signatures to xla::int64

Alex Şuhan committed 7 years ago

e4afba3f

Add in-place squeeze to ATen XLA tensor

Alex Şuhan committed 7 years ago

acedff08

Make parallel build the default.

dlibenzi committed 7 years ago

32ea2d79

Make Build{Triu, Tril} take xla::int64 diagonal

Alex Şuhan committed 7 years ago

64ea84d9

Add diagonal to ATen XLA tensor

Alex Şuhan committed 7 years ago

2755e2cc

Added end-to-end MNIST tensor training test.

dlibenzi committed 7 years ago

5dbac13d

Add pending PR patch which fixes the Python device property

Alex Şuhan committed 7 years ago

c99078fc

Be more accepting from things coming from optimizers into in-place operations.

dlibenzi committed 7 years ago

46e829e7

Handle empty*() and s_copy_() on XLA side.

dlibenzi committed 7 years ago

a98910a0

Add partial support for einsum to ATen XLA tensor

Alex Şuhan committed 7 years ago

ab15018b

The mul() operation as well might be coming with double ATEN tensor.

dlibenzi committed 7 years ago

83bd49e1

Fill in double values when lowering a scalar.

dlibenzi committed 7 years ago

a06cbc3f

Relax tolerance for TestMatmul_2x2

Alex Şuhan committed 7 years ago

cf2ddf0f

Add constant_pad_nd to ATen XLA tensor

Alex Şuhan committed 7 years ago

d14abc95

Fix Python print.

dlibenzi committed 7 years ago

07527f2f

Added utility to fetch Python frame information and to report IR Node

dlibenzi committed 7 years ago

5c991845

Added aten::fmod operation.

dlibenzi committed 7 years ago

f40f3ff4

Add at::rsqrt and at::reciprocal

vishwakftw committed 7 years ago

dc41258c

Adapt code generation to the new PR which brings per-scalar XLA type IDs.

dlibenzi committed 7 years ago

33787e26

Add binary bitwise operations to ATen XLA tensor

Alex Şuhan committed 7 years ago

06cfdaa0

Prefer device data when translating at::Tensor to XLA ones. Also allow TF to map DOUBLE type in XRT client.

dlibenzi committed 7 years ago

9ec7f63b

Make the precision a global flag

Alex Şuhan committed 7 years ago

202488ab

Relax absolute tolerance for TestLogSoftmaxBackward

Alex Şuhan committed 7 years ago

71ddc4ca

Add full precision option to matmul

Alex Şuhan committed 7 years ago

472d4767

Older