pytorch/xla

Pull Requests Commits

Fix scatter_add() for target dimension smaller than source and index.

dlibenzi committed 5 years ago

b3fedbf6

Update the skip test list (#1671)

JackCaoG committed 5 years ago

Verified 766bfefb

Skip new tests for cdist.

Ailing Zhang committed 5 years ago

7184c1e8

Use a rendezvous() on the wait of of save().

dlibenzi committed 5 years ago

d99ef16e

Have the mesh size come from configuration to allow rendezvous working with MT parallelism as well.

dlibenzi committed 5 years ago

b2368efd

Cleanup all colab notebooks to use setup script.

jysohn23 committed 5 years ago

586d54f3

Fix nightly wheel update script (#1668)

jysohn23 committed 5 years ago

Verified bd49348c

Recover from dead and hanging client-side workers (#1641)

jysohn23 committed 5 years ago

Verified b3e3d5b7

Add env-setup.py script to install and setup tpu (#1662)

jysohn23 committed 5 years ago

Verified 58cc64c0

Avoid copying TF string into std:string.

dlibenzi committed 5 years ago

540f7e97

Add test for rendezvous to the set of python tests.

dlibenzi committed 5 years ago

ec3336fa

Disable test_eig_with_eigvec which fails for precision on XLA.

dlibenzi committed 5 years ago

c0205fa5

Added user visible rendezvous interface.

dlibenzi committed 5 years ago

97f461f0

Make reductions work with dynamic shapes.

dlibenzi committed 5 years ago

b1eeb64a

Revert "Fix copy_ from xla -> cpu."

dlibenzi committed 5 years ago

6cad3b18

Disable digamma test failing on TPU.

dlibenzi committed 5 years ago

b06662c8

Skip test OOMing.

dlibenzi committed 5 years ago

a4155ce2

Add logsumexp op

JackCaoG committed 5 years ago

49c719ec

Fix bitwise ops shape inference.

dlibenzi committed 5 years ago

e2c1d5ef

Emit the tensors graph before compiling to make sure some level of debugging can be done on the TEXT graph in case of XLA lowering errors.

dlibenzi committed 5 years ago

08e4ae12

Correctly marshal the new XRT metrics.

dlibenzi committed 5 years ago

3109a2dd

Partitions the transfer to server to avoid tripping the Protocol Buffers 2GB limit.

dlibenzi committed 5 years ago

01dd6f96

Fix copy_ from xla -> cpu.

Ailing Zhang committed 5 years ago

8b389f67

Revert "Keep the memory used for sending tensors to device under control, to avoid tripping over the ProtocolBuffer 2GB time bomb."

dlibenzi committed 5 years ago

f77a9474

Keep the memory used for sending tensors to device under control, to avoid tripping over the ProtocolBuffer 2GB time bomb.

dlibenzi committed 5 years ago

575e9237

Fix copy_() operation.

dlibenzi committed 5 years ago

0210cde0

Remove unused header.

Ailing Zhang committed 5 years ago

9e05cefa

Use more XLA constants.

dlibenzi committed 5 years ago

c1b6ee90

Fix deps of googletest before the broken commit.

Ailing Zhang committed 5 years ago

ac897428

Adds multicore AlexNet on Fashion MNIST to list of README Colabs (#1637)

mruberry committed 5 years ago

Verified 2c14f650

Older