pytorch/xla

Pull Requests Commits

More VLOG around device barrier locking.

dlibenzi committed 6 years ago

89a3ad00

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

Restructure setup.py and add the GIT revision at which the library was built.

dlibenzi committed 6 years ago

d3b1701d

Change to per-file import

Ailing Zhang committed 6 years ago

da9ae4c5

Move torch_xla_py to torch_xla. Fix all uses

Ailing Zhang committed 6 years ago

3f3bfa04

Use remote handle value to make parameters unique and added check to verify that supplied parameters are matching the XLA program shape.

dlibenzi committed 6 years ago

ed502f69

Better stack trace argument dedup. (#1082)

dlibenzi committed 6 years ago

1ecceec0

Add slow XLA compile tracking and more graph hash logging. (#1078)

dlibenzi committed 6 years ago

8523c0df

Made XLA to() API support self being a CPU tensor. (#1079)

dlibenzi committed 6 years ago

71fb134a

Disable two failed tests from multiple dispatch.

Ailing Zhang committed 6 years ago

c58e186c

Fix up comment to clarify only TPU Pods be used (#1076)

jysohn23 committed 6 years ago

Verified 3e12428e

Use docker version 347. (#1071)

ailzhang committed 6 years ago

Verified 4e5dc58b

Older