pytorch/xla

Pull Requests Commits

Restructure bench framework to use classes and allow warmup runs.

dlibenzi committed 6 years ago

4aa4eaf5

Merge pull request #538 from pytorch/bench_bw

dlibenzi committed 6 years ago

Verified 01f1e46d

Merge pull request #537 from pytorch/promotem_all

dlibenzi committed 6 years ago

Verified 7fe352b8

Add benchmark with transfer in loop.

dlibenzi committed 6 years ago

524829c5

Do shape/type promotion for binary ops.

dlibenzi committed 6 years ago

d694e313

More concise error message in CheckIntList

asuhan committed 6 years ago

88736aa2

Clean up lambda captures in pooling operations

asuhan committed 6 years ago

93956a9d

Resize kernel, stride and padding for pooling operations

asuhan committed 6 years ago

5c498c90

Support for n-dimensional max pooling

asuhan committed 6 years ago

13bbaa06

Support for n-dimensional average pooling

asuhan committed 6 years ago

862aceb7

Merge pull request #532 from pytorch/add_py_bench

dlibenzi committed 6 years ago

Verified 2509d5e7

Added Python benchmark to measure Cloud TPU latency testing.

dlibenzi committed 6 years ago

02a01396

Fix the XLA device guard to add noexcept qualifier.

dlibenzi committed 6 years ago

9e7a8239

Fix Colab to use JIT MNIST as the XTen version is not yet supported in the wheels we can supply for TF 1.13 compatibility.

dlibenzi committed 6 years ago

3b8bda4c

Fixes zeros_like() monkey-patch in XLA Python init.

dlibenzi committed 6 years ago

2ffe5a69

Support direct remap ATEN generation to be used for TH functions (and similar other scenarios).

dlibenzi committed 6 years ago

b567bbdb

Count XLA tensors.

dlibenzi committed 6 years ago

c79202dd

Bump compilation caches up and made them environment configurable.

dlibenzi committed 6 years ago

33358305

Fix selected slice shape.

dlibenzi committed 6 years ago

08f8d6f8

Bound the step for slices.

dlibenzi committed 6 years ago

0e9c6f6c

Remove TensorData and use DeviceData back as it is no more needed.

dlibenzi committed 6 years ago

a4b7006e

Added Select and Unselect operations to handle slices with strides != 1.

dlibenzi committed 6 years ago

eb316286

Enable the model comparator to be left active and unconfigured, resulting in noop.

dlibenzi committed 6 years ago

470c9444

Added aten::triangular_solve op

vishwakftw committed 6 years ago

3e5051fb

Fix XLA tensor views which are not handled by the XLA tensor core.

dlibenzi committed 6 years ago

3ff502ba

Minor cleanup of model comparator.

dlibenzi committed 6 years ago

aae5e5cf

Added model comparator Python module.

dlibenzi committed 6 years ago

1dfe3e2b

Fix aten::clamp to use lowest() instead of min().

dlibenzi committed 6 years ago

7746ff65

Replace IntList with IntArrayRef

asuhan committed 6 years ago

90849b0f

Add index_{add, copy} and their inplace versions to ATen XLA tensor

asuhan committed 6 years ago

78a09d93

Older