pytorch/xla

Pull Requests Commits

Build GCS file system support.

dlibenzi committed 6 years ago

a408a4fc

Add overrides for native_layer_norm and its backward.

Ailing Zhang committed 6 years ago

d9977c6e

Let the caller handle the separate paths to be fed to the xm.save() API in case master_only is False.

dlibenzi committed 6 years ago

531a26f2

Add helper function to get Hlo graph in test.

Ailing Zhang committed 6 years ago

6bacd3ad

Added TFRecord Example reader interface.

dlibenzi committed 6 years ago

0d4c160f

Allow multi-processing POD to recognize local devices.

dlibenzi committed 6 years ago

3e62c8bb

Add missing instructions to the README

sprt committed 6 years ago

ee83d05b

Tweak README to make it more beginner-friendly

sprt committed 6 years ago

12a9b29e

Add API to wait for pending async ops. Make OpByOp use a hash key seed based on replication devices.

dlibenzi committed 6 years ago

4d2c8746

Fix clang-format cmd typo

jysohn23 committed 6 years ago

a15bd007

Log thread IDs by default.

dlibenzi committed 6 years ago

ba523346

Added mse_loss and mse_loss_backward lowering. (#1206)

ailzhang committed 6 years ago

Verified 950e12a6

Moved reduction mapping API to tensor_methods.cpp

dlibenzi committed 6 years ago

de07b24d

Vlog the client mesh retrieved from client mesh master.

dlibenzi committed 6 years ago

348676e9

Fix empty env_var crash for xla_dist

jysohn23 committed 6 years ago

7a466971

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

ailzhang committed 6 years ago

Verified ea7767f9

Flush stderr/stdout before exiting due to error.

jysohn23 committed 6 years ago

28581a65

Added MarkStep counter.

dlibenzi committed 6 years ago

b6c59a36

Wrapping namespace Reduction in namespace at (#1176)

yf225 committed 6 years ago

4d6f7da9

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

dlibenzi committed 6 years ago

864acbcf

Disable op-by-op tests until issue fixed.

dlibenzi committed 6 years ago

de3e16f8

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

dlibenzi committed 6 years ago

1f8f8aea

Clarify ParallelLoader API.

dlibenzi committed 6 years ago

e1904879

fixed a typo for graph dumping

Min Xu committed 6 years ago

12952695

Add explicit enumerate on Colab notebooks (#1190)

jysohn23 committed 6 years ago

Verified 9ae81316

Restructure and add more logging.

dlibenzi committed 6 years ago

4cf447bb

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

dlibenzi committed 6 years ago

95879f3c

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

dlibenzi committed 6 years ago

d8c47b61

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

dlibenzi committed 6 years ago

2244123d

Remove the enumerate-like parallel loader iterator behavior.

dlibenzi committed 6 years ago

888abc54

Older