pytorch/xla

Pull Requests Commits

If a tensor is already CPU, should be passed as is out of the conversion API.

dlibenzi committed 6 years ago

398fb9e0

Special case ones() and zeros() APIs.

dlibenzi committed 6 years ago

9bcccecc

Register float as last type.

dlibenzi committed 6 years ago

62c17e14

Let the generator see Tensor copy APIs.

dlibenzi committed 6 years ago

a17ea357

Remove PyTorch patches

asuhan committed 6 years ago

f1e93902

Rename IR Node to IR Value to fit the new data structure name.

dlibenzi committed 6 years ago

6b0d8743

Fix aten_xla_bridge.{h, cpp} formatting

asuhan committed 6 years ago

d73ff2bd

Allow shape to be passed to NotSupportedOp

asuhan committed 6 years ago

97338749

Replace placeholder max_pool2d_indices with not supported node

asuhan committed 6 years ago

a50608ab

Replace remaining make_shared with ir::MakeNode

asuhan committed 6 years ago

20ae3cd1

Add max_pool2d_with_indices_backward to ATen XLA tensor

asuhan committed 6 years ago

440919d6

Add MaxPool2dIndices placeholder

asuhan committed 6 years ago

95dd9520

Fix build after recent PyTorch core changes

asuhan committed 6 years ago

35c7ae78

Drop the commit by leaving open the door for future uses.

dlibenzi committed 6 years ago

9cb517a8

Wire in ATEN XLA Python bindings and added simple end-to-end test.

dlibenzi committed 6 years ago

bf22f0fd

A few more replacements.

dlibenzi committed 6 years ago

c79bf256

Apply some de-auto refactoring.

dlibenzi committed 6 years ago

e1d73823

Fix the operator kind in NllLossBackwardOp

asuhan committed 6 years ago

ab187764

Fix build after latest PyTorch changes

asuhan committed 6 years ago

02ae0f71

Fix the XLA tensor get-xla handling, and bump up the size of conv2d bkw test.

dlibenzi committed 6 years ago

af1c99fa

Add ir::MakeNode<>() utility API to avoid casting IR Nodes into Values.

dlibenzi committed 6 years ago

f6dd3118

Rename NodeOperand to Value.

dlibenzi committed 6 years ago

9eef3b57

Adjust ATen XLA tests for cloud TPU

asuhan committed 6 years ago

da0e3585

Remove explicit ir::NodeOperand calls where possible

asuhan committed 6 years ago

f82bedcc

Add nll_loss_backward to ATen XLA tensor

asuhan committed 6 years ago

032876b9

Add nll_loss to ATen XLA tensor

asuhan committed 6 years ago

11c8465e

Add mm to ATen XLA tensor

asuhan committed 6 years ago

eef47fe2

Add transpose test

asuhan committed 6 years ago

062f5d38

Add threshold_backward to ATen XLA tensor

asuhan committed 6 years ago

68dc42ce

Add _log_softmax_backward_data to ATen XLA tensor

asuhan committed 6 years ago

420f26e9

Older