pytorch/xla

Pull Requests Commits

Add function to detect tensor potential aliases.

dlibenzi committed 6 years ago

07318a3f

Added more meaningful device spec print for multi-processing.

dlibenzi committed 6 years ago

68f9db8b

Added MP API documentation.

dlibenzi committed 6 years ago

45b9e4a2

add issue templates (#1036)

ailzhang committed 6 years ago

Verified 7d0abbc7

Use static for auto-unregistration object.

dlibenzi committed 6 years ago

429c89ad

draft contribution guide for discussion (#1033)

ailzhang committed 6 years ago

Verified fb6b0788

Fixed the XLA _out() code generation and added a few more overrides.

dlibenzi committed 6 years ago

040166bc

Ailing Zhang committed 6 years ago

a7e47149

Ailing Zhang committed 6 years ago

e672f3ec

Allow reduce operations in every point of the graph.

dlibenzi committed 6 years ago

acff0b9a

allow update any env (#1028)

taylanbil committed 6 years ago

Verified 8bc314b2

Added XLA multi-processing support and API.

dlibenzi committed 6 years ago

1f45f0a3

Make add_scatter_combiner a std::function

asuhan committed 6 years ago

271c6fcf

allow copy.copy() to return xla tensor (#1018)

ailzhang committed 6 years ago

Verified 2e67d89a

Fix conv backward hash population and ToString formatting.

dlibenzi committed 6 years ago

2ba4ec4c

Fix convolution ToString() formatting.

dlibenzi committed 6 years ago

856b864a

Override is_floating_point() and is_signed().

dlibenzi committed 6 years ago

da89b09c

Wire in dense scatter.

dlibenzi committed 6 years ago

a1c0c5fe

Migrate to new TF version and added test for 0dim index_select().

dlibenzi committed 6 years ago

a8d9bc1b

fix pred promotion for arithmetic ops (#1009)

ailzhang committed 6 years ago

Verified 43d35cd9

Fix test to use proper scalar type.

dlibenzi committed 6 years ago

62282568

stop promoting from pred to u8 for bitwise ops (#986)

ailzhang committed 6 years ago

Verified d36f3b4a

Allow passing generic args to the optimizer step.

dlibenzi committed 6 years ago

91fc0550

Make the topology fetch API do not require the XRT computation client to be setup.

dlibenzi committed 6 years ago

cd8743c1

punt back to cpu for xla::dot related ops with int input (#999)

ailzhang committed 6 years ago

a98273a6

add comment about half (#998)

ailzhang committed 6 years ago

Verified 3cb6e972

better error message for arange (#1002)

ailzhang committed 6 years ago

9f6a18ee

Make sign and abs work with unsigned integers

asuhan committed 6 years ago

45507ac6

Shard datasets for distributed training. (#997)

jysohn23 committed 6 years ago

Verified 6341b5c0

add global rate and wall time to train print statement (#988)

zcain117 committed 6 years ago

Verified 582f612f

Older