pytorch/xla

Pull Requests Commits

Move the metrics report into a dedicated xla_model API.

dlibenzi committed 6 years ago

73fc902f

Remove serialization patch & add a few more tests. (#1067)

ailzhang committed 6 years ago

Verified b91cf83b

Clean up unnecessary includes and restructure API condition flow. (#1121)

dlibenzi committed 6 years ago

aa4cd2cd

Add test_bool_sub to allowed tests. (#1118)

ailzhang committed 6 years ago

Verified b2c31918

Remove unused argument.

dlibenzi committed 6 years ago

36edb547

Set high precision even for XLA tensor tests.

dlibenzi committed 6 years ago

47f778f4

Show build time GIT revision at library init.

dlibenzi committed 6 years ago

5bb7d9b5

Moving the project into a team effort instead of individuals.

dlibenzi committed 6 years ago

be9219e1

More VLOG around device barrier locking.

dlibenzi committed 6 years ago

483298ab

Re-title Best Practice to Best Practices (#1112)

taylanbil committed 6 years ago

993090b4

Excludes many_batches tests and tests failing on TPUs (#1111)

mruberry committed 6 years ago

Verified df18d1b7

Add more VLOG around compile/execute ops.

dlibenzi committed 6 years ago

4dac2e24

adds programmatic test device support and test_indexing.py

Mike Ruberry committed 6 years ago

74cff13f

Doc improvements suggested in the meeting. (#1107)

ailzhang committed 6 years ago

Verified 7ad70181

switch resnet50 to default hyperparams that gave 76.27% final accuracy (#1103)

zcain117 committed 6 years ago

Verified 67a0c035

Cleanup TROUBLESHOOTING.md a bit (#1105)

asuhan committed 6 years ago

bceac473

Add the pytorch GIT revision as well.

dlibenzi committed 6 years ago

bd868b3a

Move building manually out of Readme (#1101)

ailzhang committed 6 years ago

Verified 2c02279f

Refactor docs. (#1095)

ailzhang committed 6 years ago

Verified fe7f5942

Added small section documenting the stack trace collection.

dlibenzi committed 6 years ago

edfe84ff

Fix hash collision issue.

dlibenzi committed 6 years ago

76cf5668

Enable test_neg on XLA. (#1091)

ailzhang committed 6 years ago

Verified e3d37493

Added utility script to dump GDB stacks.

dlibenzi committed 6 years ago

cc4c9af9

Stop parsing NativeFunctions.h since we know enough from Functions.h (#1094)

ailzhang committed 6 years ago

Verified 2943bf2d

Fix convolution tests precision to fit TPU, and also print the differences of the compared inputs.

dlibenzi committed 6 years ago

0004b92e

Drop the unwrap business and handle any sort of tensor type coming in where they should have not.

dlibenzi committed 6 years ago

321e3197

Removes several tests that were not properly tested on XLA (#1090)

mruberry committed 6 years ago

Verified dccb91fe

Make patch to fail in case of conflicts.

dlibenzi committed 6 years ago

e8b1ad78

Fix torch_patch as broken by PT#26339 (#1088)

ailzhang committed 6 years ago

Verified daa9e435

allow partial warmup epochs in learning rate scheduler (#1087)

zcain117 committed 6 years ago

Verified a84e0725

Older