pytorch/xla

Pull Requests Commits

Enabled booting from a slice bigger than a donut.

dlibenzi committed 6 years ago

b3c660d6

Add script to install torch and torch_xla from source (#708)

jysohn23 committed 6 years ago

Verified b46e2ade

Fix the daily breakage.

dlibenzi committed 6 years ago

1b5c22da

Fix Variable::data() to Variable::tensor_data() API change.

dlibenzi committed 6 years ago

89b06a12

Remove landed 20783 PR.

dlibenzi committed 6 years ago

fbc76442

Mention Nightly builds in README.

dlibenzi committed 6 years ago

03d12b31

Populate HLO metadata if XLA_HLO_DEBUG is set.

dlibenzi committed 6 years ago

ab90f596

Move HLO generation to IR dump util.

dlibenzi committed 6 years ago

90d3c2b4

Update the README file.

dlibenzi committed 6 years ago

84239a2b

Add support for pooling tensors without batch

asuhan committed 6 years ago

15a8dc62

Add incoming PT PR to fix the s_copy premature removal.

dlibenzi committed 6 years ago

7bd0debf

Fix scalar_tensor for XLA backend

asuhan committed 6 years ago

317ee743

Add scatter_add to ATen XLA tensor

asuhan committed 6 years ago

51ac589a

Fix removal of ATEN s_copy* interfaces.

dlibenzi committed 6 years ago

8020a932

Add benchmark for tensor send to backend.

dlibenzi committed 6 years ago

481d5b5a

Drop diplicated import and reorder.

dlibenzi committed 6 years ago

5af84c34

Avoid repeating adding the optimizer's state.

dlibenzi committed 6 years ago

3abdf335

Make sure the copy partition size is never 0. This could have happened, for example, in a machine with many cores (ie, > 64), and a tensor of shape (32, 32, 32, 32, 32).

dlibenzi committed 6 years ago

9fe779a7

Added option to save HLO test on file, in case of error, as log lines gets intermingled.

dlibenzi committed 6 years ago

2755abd3

Dump the XLA HLO graphs in case of compile errors.

dlibenzi committed 6 years ago

f1f57f31

Do not print the loss at every step.

dlibenzi committed 6 years ago

328ccddf

Fix threaded copy code to avoid selecting the strided copy dimension as threading dimension.

dlibenzi committed 6 years ago

3addbe8f

Slice a tensor copy over multiple threads.

dlibenzi committed 6 years ago

0b1e034d

Tune hyperparams for ResNet18 on cifar

jysohn23 committed 6 years ago

56d53b5d

Allow hyper parameter tuning and scale LR for MNIST multi core. (#682)

dlibenzi committed 6 years ago

69c50145

Script to update torch and torch_xla installation (#681)

jysohn23 committed 6 years ago

Verified 0848d793

Fix build breakage due to API changes.

dlibenzi committed 6 years ago

37363b46

Fix PyTorch Tensor virtual function signature change.

dlibenzi committed 6 years ago

698eec31

Collect torch build artifact

jysohn23 committed 6 years ago

Verified 04353988

Add support for max and min across dimensions

asuhan committed 6 years ago

7cb7ba04

Older