Commits

Branch selector

 master

User selector

blakehechtman

Datepicker

All time

Commit History

Commits on Apr 24, 2019

• [XLA] Fix dot strength reduction to simply remove all degenerate dimensions and

  Show description for 86fbd13
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  86fbd13

  Copy full SHA for 86fbd13

Commits on Apr 23, 2019

• [TF:SPECIAL_MATH_OPS] Call xla_einsum in supported cases.

  Show description for c74c1a2
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  c74c1a2

  Copy full SHA for c74c1a2

Commits on Apr 20, 2019

• [TF2XLA] Use Two pass argmax on CPU as well for CatagoricalOp.

  Show description for 8763339
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  8763339

  Copy full SHA for 8763339

Commits on Apr 18, 2019

• [XLA] Use variadic reduce for argmin/argmax implementation to avoid issues with

  Show description for 94745ad
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  94745ad

  Copy full SHA for 94745ad

Commits on Apr 10, 2019

• Enable broadcasting in TF2XLA MatMul.

  Show description for b41044a
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  b41044a

  Copy full SHA for b41044a

Commits on Apr 2, 2019

• [XLA:CLIENT] Support numpy style implicit broadcasting in xla::BatchDot

  Show description for 6f6740c
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  6f6740c

  Copy full SHA for 6f6740c

Commits on Apr 1, 2019

• Propagate the layout of reshapes from rank 1 depth first.

  Show description for 9fd90ad
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  9fd90ad

  Copy full SHA for 9fd90ad

Commits on Mar 29, 2019

• [XLA] Add an environment variable that allows xla to increase the precision inside of fusion instructions.

  Show description for d11a7f8
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  d11a7f8

  Copy full SHA for d11a7f8

Commits on Mar 28, 2019

• [TF2XLA] Expose replica ids to python.

  Show description for 39a1175
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  39a1175

  Copy full SHA for 39a1175

Commits on Mar 20, 2019

• [XLA] Don't check for equality of shapes in conditional simplification incase

  Show description for 872eb47
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  872eb47

  Copy full SHA for 872eb47

• [XLA] Correct the handling of conditional instructions where operands of the

  Show description for 78d806c
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  78d806c

  Copy full SHA for 78d806c

Commits on Mar 5, 2019

• [XLA] Do not try to force a particular layout when converting from an Effective

  Show description for 50b87be
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  50b87be

  Copy full SHA for 50b87be

Commits on Mar 1, 2019

• [XLA] Propagate the layout of conditionals from the branch with the most

  Show description for e4e3de2
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  e4e3de2

  Copy full SHA for e4e3de2

• [TF2XLA] Use a single gather for reverse sequence op.

  Show description for 7391e2b
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  7391e2b

  Copy full SHA for 7391e2b

Commits on Feb 28, 2019

• [XLA:CLIENT] implement torch.index_select

  Show description for d0b4936
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  d0b4936

  Copy full SHA for d0b4936

Commits on Feb 27, 2019

• Automated rollback of commit 4bf9ea4

  Show description for 4466bff
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  4466bff

  Copy full SHA for 4466bff

• [XLA] Simplify Slice of Broadcast when only non-broadasted dimensions are

  Show description for 8e23a58
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  8e23a58

  Copy full SHA for 8e23a58

Commits on Feb 22, 2019

• [XLA] Just use ShapeUtil::AlignLayouts for reshape layout selection.

  Show description for 4bf9ea4
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  4bf9ea4

  Copy full SHA for 4bf9ea4

Commits on Feb 15, 2019

• [XLA] simplify ShapeUtil::Alignlayouts by removing degenerate dimensions.

  Show description for 40cd7f9
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  40cd7f9

  Copy full SHA for 40cd7f9

• [XLA:CLIENT] Implement a varient of gather equivalent to torch.gather

  Show description for 47694b7
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  47694b7

  Copy full SHA for 47694b7

Commits on Feb 14, 2019

• [XLA] Simplify dot operations that have only batch and contracting dimension on the lhs or rhs.

  Show description for e9129a7
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  e9129a7

  Copy full SHA for e9129a7

Commits on Feb 12, 2019

• [XLA] Simplify batch dots that have no contracting dimensions into multiplies.

  Show description for ba941c2
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  ba941c2

  Copy full SHA for ba941c2

Commits on Feb 7, 2019

• [XLA] Support scalar dots and dots without contracting dimensions general with a multiply.

  Show description for e8a82ff
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  e8a82ff

  Copy full SHA for e8a82ff

Commits on Jan 22, 2019

• [XLA:CLIENT] Make a simple einsum interface to dot general.

  Show description for f5fc7af
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  f5fc7af

  Copy full SHA for f5fc7af

Commits on Jan 18, 2019

• [XLA] Support no dimension numbers set in Dot.

  Show description for a45e725
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  a45e725

  Copy full SHA for a45e725

• [TF] Make behavior of dropout backward compatability.

  Show description for 04e8de0
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  04e8de0

  Copy full SHA for 04e8de0

Commits on Jan 16, 2019

• [TF] Use compare and cast instead of floor for dropout.

  Show description for f98110b
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  f98110b

  Copy full SHA for f98110b

Commits on Jan 15, 2019

• [XLA] Recurse through copies and reshapes to find bitcasts. Layout assignment may create an alignment of a reshape that eventually introduces unnecessary copies since there are many valid alignment…

  Show description for 92e0eea
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  92e0eea

  Copy full SHA for 92e0eea

Commits on Jan 9, 2019

• [XLA] Fix Dot operation semantics.

  Show description for 65cfc6a
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  65cfc6a

  Copy full SHA for 65cfc6a

Commits on Jan 8, 2019

• [XLA] Canonicalize dot dimension numbers on CPU and GPU backends cases that previously failed shape inference.

  Show description for a3d639a
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  a3d639a

  Copy full SHA for a3d639a

Commits on Jan 3, 2019

• Fix typo.

  Show description for 4dc0ae2
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  4dc0ae2

  Copy full SHA for 4dc0ae2

• [XLA] Propagate the layout of transposes and reshapes that add one sized

  Show description for d04b09d
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  d04b09d

  Copy full SHA for d04b09d

Commits on Dec 23, 2018

• [XLA] Allow non-monotonic batch dimensions, they are supported by the

  Show description for 6e304fa
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  6e304fa

  Copy full SHA for 6e304fa

Commits on Dec 22, 2018

• Allow dot to have mutiple contracting dimensions and non contracting dimensions.

  Show description for fa76858
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  fa76858

  Copy full SHA for fa76858

Commits on Dec 18, 2018

• [XLA] Convert reduce windows that are actually reduces into reduce and reshape.

  Show description for a4bb285
  

  blakehechtman

  authored and

  tensorflower-gardener

  committed
  a4bb285

  Copy full SHA for a4bb285

Pagination

Previous

Next