pytorch/xla

Pull Requests Commits

Add metrics graph plots to debug_run.py. (#1450)

dlibenzi committed 6 years ago

d161e213

Remove patch as the upstream PR #28493 is merged. (#1446)

ailzhang committed 6 years ago

Verified e49ac485

Print the XLA computation output shapes and layouts on error.

dlibenzi committed 6 years ago

8fb9ab06

Check reference cycles when iterating over python data structures.

dlibenzi committed 6 years ago

9785d025

Update issue template to include debug_run reports

dlibenzi committed 6 years ago

f2dc8855

Add neural style transfer notebook.

Ailing Zhang committed 6 years ago

cd465715

Append final slah on GCS list, and decode URL encoded paths.

dlibenzi committed 6 years ago

ec87d326

Test max accuracy instead of final accuracy (#1438)

zcain117 committed 6 years ago

Verified c4f507c1

Metrics comparator (#1402)

zcain117 committed 6 years ago

Verified 438f8429

Added debug_run.py section is docs and done some tweaks on it.

dlibenzi committed 6 years ago

ae9f5402

Allow metrics percentiles to be configurable.

dlibenzi committed 6 years ago

72588056

Do not add extra .0 if running single core. (#1437)

dlibenzi committed 6 years ago

b5ac7b12

Enable some dynamic shapes tests. (#1434)

dlibenzi committed 6 years ago

f66bf943

Properly namespace TorchGather(). (#1436)

dlibenzi committed 6 years ago

66fa4cd1

Fix single-core multi-processing.

dlibenzi committed 6 years ago

59ce59c7

Skip test on two logical ops

xuhdev committed 6 years ago

08cbd7af

Added debug script to run pytorch/xla models in supervised mode.

dlibenzi committed 6 years ago

77cce413

Allow parsing all graph types, not only TEXT.

dlibenzi committed 6 years ago

0a2728f3

Fix one-hot lowering. (#1425)

dlibenzi committed 6 years ago

d39fe2a8

Re-enable op-by-op test. (#1423)

dlibenzi committed 6 years ago

842b3abe

Bring new TF with AllReduce tuple fix.

dlibenzi committed 6 years ago

7ffd4d5e

Fake Token using scalar until XLA AllReduce() Token support is added.

dlibenzi committed 6 years ago

ec28190d

Make trim graph locations deterministic.

dlibenzi committed 6 years ago

8d2ae7c9

Fix the test by avoiding the current (will be fixed soon) XLA AllReduce bug on tuple inputs with single element inside.

dlibenzi committed 6 years ago

ea39a80a

Fix MP AllReduce test.

dlibenzi committed 6 years ago

93f6d209

Do not spawn a different process if we will be running on one single core.

dlibenzi committed 6 years ago

04944651

Allow pow() to take int exponents as the lowering does the promotions already.

dlibenzi committed 6 years ago

6e436eb4

Temporarily disable op-by-op test.

dlibenzi committed 6 years ago

0bd9a898

Use shape_fn for expensive shape calculations into GenericOp(). (#1410)

dlibenzi committed 6 years ago

954c99a2

Added aten::reflection_pad2d and aten::reflection_pad2d_backwar… (#1409)

dlibenzi committed 6 years ago

012367f6

Newer Older