pytorch/xla

Pull Requests Commits

Make for_each API deterministic.

dlibenzi committed 5 years ago

94771004

Vlog the client mesh retrieved from client mesh master.

dlibenzi committed 5 years ago

348676e9

Fix empty env_var crash for xla_dist

jysohn23 committed 5 years ago

7a466971

Add memory_format to `to` overrides. (#1200)

ailzhang committed 5 years ago

Verified ea7767f9

Flush stderr/stdout before exiting due to error.

jysohn23 committed 5 years ago

28581a65

Added MarkStep counter.

dlibenzi committed 5 years ago

b6c59a36

Wrapping namespace Reduction in namespace at (#1176)

yf225 committed 5 years ago

4d6f7da9

Introduce local ordinal API to allow restricting actions only on each host master device (0).

dlibenzi committed 5 years ago

864acbcf

Disable op-by-op tests until issue fixed.

dlibenzi committed 5 years ago

de3e16f8

Added xm.save() utility API to save the model's state without incurring into

dlibenzi committed 5 years ago

1f8f8aea

Clarify ParallelLoader API.

dlibenzi committed 5 years ago

e1904879

fixed a typo for graph dumping

Min Xu committed 5 years ago

12952695

Add explicit enumerate on Colab notebooks (#1190)

jysohn23 committed 5 years ago

Verified 9ae81316

Restructure and add more logging.

dlibenzi committed 5 years ago

4cf447bb

Make comparison ops use device data in place of in-graph scalars, to reduce recompilations.

dlibenzi committed 5 years ago

95879f3c

Added more debugging info in the core tensor code. (#1180)

dlibenzi committed 5 years ago

d8c47b61

Added aten::l1_loss and aten::l1_loss_backward support. (#1183)

dlibenzi committed 5 years ago

2244123d

Remove the enumerate-like parallel loader iterator behavior.

dlibenzi committed 5 years ago

888abc54

Add links to Cloud Tutorials (#1172)

jysohn23 committed 5 years ago

Verified 1ef5dfcf

Remove override of Tensor::numel since this goes directly to TensorImpl one. (#1175)

ailzhang committed 5 years ago

Verified 7f70ab75

Do not force layout reorder in case the padding amount is within XLA_MAX_PADDING_FACTOR range (default 25%).

dlibenzi committed 5 years ago

60dbe563

Fix API comments.

dlibenzi committed 5 years ago

152f3db1

No need to fetch gradients in non replicated mode.

dlibenzi committed 5 years ago

b70ebb93

Use torch_xla.debug.metrics for notebooks (#1170)

jysohn23 committed 5 years ago

Verified 4eb4f92d

Add missing TPU check cell on ResNet18 colab (#1169)

jysohn23 committed 5 years ago

Verified 54159f9b

Update colab notebooks (#1168)

jysohn23 committed 5 years ago

Verified b678e354

Do not error out if sync of empty tensors set is combined with tensor.cpp VLOG.

dlibenzi committed 5 years ago

d7a7495e

no .nojekyll (#1167)

mruberry committed 5 years ago

Verified bcc57b50

nojekyll (#1165)

mruberry committed 5 years ago

Verified 6a558db3

Revert "Remove override of Tensor::numel since this goes directly to TensorImpl one. (#1162)" (#1164)

dlibenzi committed 5 years ago

Verified 050b24cd

Older