pytorch/xla

Pull Requests Commits

Given that we can create our own symbols using their API, use c10::Symbol instead of strings.

dlibenzi committed 6 years ago

4b907b6f

Update the PyTorch commit id to latest

asuhan committed 6 years ago

e0ad9bfe

Replace big translator switch with map of handlers

asuhan committed 6 years ago

5fb376b8

Ailing Zhang committed 6 years ago

d9ecd4ec

Avoid rebuilding if there is no need to.

dlibenzi committed 6 years ago

efa2e46b

Sketch IR interface.

dlibenzi committed 6 years ago

eb966ffb

Remove stale TODO

asuhan committed 6 years ago

b5ac3170

Update the PyTorch commit id to latest

asuhan committed 6 years ago

7483dd04

Fix ReplaceInPlaceOps pass

asuhan committed 6 years ago

1f94a9c6

Apply clang-format to all files

asuhan committed 6 years ago

7238d80f

Move our code out of torch::jit into torch_xla.

dlibenzi committed 6 years ago

9f0ddc82

Remove stale TODO

asuhan committed 6 years ago

a7313a3f

Add support for simple cases of adaptive average pool

asuhan committed 6 years ago

ca68bbae

Replace use of deprecated xla::ShapeUtil methods

asuhan committed 6 years ago

3f110236

Remove hard coded version from installation name

jysohn23 committed 6 years ago

Verified 711f3087

jysohn23 committed 6 years ago

dd10b00d

Remove hard coded version from installation name

jysohn23 committed 6 years ago

1436e123

Remove hard coded version from installation name

jysohn23 committed 6 years ago

c7c8ccbc

Change torch version

jysohn23 committed 6 years ago

35a09de8

Fix passes/replace_untraced_operators.cpp formatting

asuhan committed 6 years ago

8347b7a2

Support constant bool list in the translator

asuhan committed 6 years ago

edc33348

Use the new XRT feature that allows to release multiple handles with a single TF node.

dlibenzi committed 6 years ago

cae5940b

Remove stray return statements.

dlibenzi committed 6 years ago

941ba922

Remove the expand pass and implement the SumToSize operation.

dlibenzi committed 6 years ago

f2cab69c

Actually promote the results of trivial binary operations.

dlibenzi committed 6 years ago

ee87071f

Add addcdiv_ and addcmul_ required for certain PT optimizers (ie, adam).

dlibenzi committed 6 years ago

965deff2

Support elementwise division and promote shapes as well. Also support sqrt and rsqrt.

dlibenzi committed 6 years ago

e98572aa

Uniform error handling.

dlibenzi committed 6 years ago

83f22df7

Simplify client code.

dlibenzi committed 6 years ago

4a0b5b09

Refactor the translator error handing to the "XLA" one, now that we have a CHECK-less error reporting.

dlibenzi committed 6 years ago

d7bec1a3

Newer Older