pytorch/xla

Pull Requests Commits

Minor cleanup of model comparator.

dlibenzi committed 6 years ago

b673dcbc

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

No need to add CRS for one device replications.

dlibenzi committed 6 years ago

963e6fa6

Added script to normalize the IR graph text output, to allow textual diffs.

dlibenzi committed 6 years ago

5686da1d

Enable shape caching for IndexOp

asuhan committed 6 years ago

d8be8bd9

Replace Shape::dimensions_size() uses with Shape::rank()

asuhan committed 6 years ago

717db87f

Add _unsafe_view to ATen XLA tensor

asuhan committed 6 years ago

aa597be8

Pin MKL to older version as the new one which shipos with Conda breaks the PyTorch build.

dlibenzi committed 6 years ago

4455e8ef

Add index_fill to ATen XLA tensor

asuhan committed 6 years ago

d19ba5d2

Avoid trimming IR graph during sync.

dlibenzi committed 6 years ago

ea6bda64

Fix issues in https://github.com/pytorch/xla/issues/501 related to PT operations which should have been views into the buffers.

dlibenzi committed 6 years ago

079d1240

Use a global apply tensors graph cache, to boost callers other than module.cpp.

dlibenzi committed 6 years ago

fb922759

Add support to auto-generate *_out() functions.

dlibenzi committed 6 years ago

5e37ec43

Fix gcc 7.3 build

asuhan committed 6 years ago

c79f9dad

Add index_put, index_put_ to ATen XLA tensor

asuhan committed 6 years ago

f6fc28ad

Remove landed 17847 PR.

dlibenzi committed 6 years ago

55f74f0e

Fix changes in PyTorch argmax/argmin APIs.

dlibenzi committed 6 years ago

d684efe2

Make IR graph trim parameters configurable.

dlibenzi committed 6 years ago

9478f472

Fix std::string hash.

dlibenzi committed 6 years ago

17951fab

Add shape caching to more IR ops.

dlibenzi committed 6 years ago

25c4064a

Bridge in autograd PR.

dlibenzi committed 6 years ago

bf8a1770

Avoid expensive shape computations while building IR nodes.

dlibenzi committed 6 years ago

9085c8f6

Drop changes tools/build_pytorch_libs.py which landed

asuhan committed 6 years ago

ff09384e

Simplify an open coded loop.

dlibenzi committed 6 years ago

fbbf2f58

Add window factories to ATen XLA tensor

asuhan committed 6 years ago

df327ef9

Add pixel_shuffle to ATen XLA tensor

asuhan committed 6 years ago

9d376f33

Add sum_to_size to ATen XLA tensor

asuhan committed 6 years ago

a089e862

Older