pytorch/xla

Pull Requests Commits

Always lower to masked_scatter() as it does not produce dynamic outputs.

dlibenzi committed 5 years ago

ff0cdd41

Do not reshape to R1 the masked scatter input.

dlibenzi committed 5 years ago

3838c4d9

Added masked_scatter support.

dlibenzi committed 5 years ago

f0552723

Fixup grad clip patch.

dlibenzi committed 5 years ago

b6fad06d

Rename counter to the new name uses in the other functions.

dlibenzi committed 5 years ago

78c5c5df

Cope with __or__ and __ior__ overrides going away.

dlibenzi committed 5 years ago

d39c9490

Pass source shapes instead of raw dimensions, down to views.

dlibenzi committed 5 years ago

d62a5756

Fix Python test.

dlibenzi committed 5 years ago

730b50fd

Some more shape cleanup.

dlibenzi committed 5 years ago

1eacf89a

Cope with AND overrides gone.

dlibenzi committed 5 years ago

d6ad73cb

Make more dynamic shapes propagation.

dlibenzi committed 5 years ago

76598cf0

Measure device lock time.

dlibenzi committed 5 years ago

dd9065ed

Add metric to track single transfers of tensors to device.

dlibenzi committed 5 years ago

5d18faf0

Fix the upsample* APIs to take optionals.

dlibenzi committed 5 years ago

b67c7f00

No need to compute layouts for normal shape() API.

dlibenzi committed 5 years ago

b9ade635

Re-enable test after XLA fixes.

dlibenzi committed 5 years ago

b016d3ec

Some code refactoring and removal of unnecessary include file.

dlibenzi committed 5 years ago

225483b5

Bring in new TF with dynamic concat shape inference fixed.

dlibenzi committed 5 years ago

9ca7da11

Fix symbol of Resize node

asuhan committed 5 years ago

296f111d

Optionally dump graphs in C++ tests.

dlibenzi committed 5 years ago

e6364c4d

Cleanup test runner and disable currently failing dynamic tests. (#1502)

dlibenzi committed 5 years ago

e0e80a44

Bring in new TF.

dlibenzi committed 5 years ago

0fd9955a

Expand more dynamic reshapes coverage.

dlibenzi committed 5 years ago

c487f462

Use bridge APIs in XLA tensor impl.

dlibenzi committed 5 years ago

33bd8949

Expand dynamic dimension propagation.

dlibenzi committed 5 years ago

7cad5347

Use a common source of current device.

dlibenzi committed 5 years ago

b29de375

Change Bazel version in docs.

dlibenzi committed 5 years ago

455a15a2

Refine the logic of assigning the output dynamic dimension of a reshape.

dlibenzi committed 5 years ago

415ea731

Switch to tensorboardX (#1492)

zcain117 committed 5 years ago

Verified a024e54d

Removes GDrive requirement, style updates

mruberry committed 5 years ago

bec9c0e3

Older