Code

update#19

Code

Merged

AnnaTrainingG merged 25 commits intoAnnaTrainingG:developAnnaTrainingG/Paddle:developfrom PaddlePaddle:developPaddlePaddle/Paddle:developCopy head branch name to clipboardJul 6, 2021

+2,257-578

Commits

Commits on Jul 5, 2021

• fix decoding error when clip grad op and python==2 (#33937)

  jiweibo

  authored
  c661783

  Copy full SHA for c661783

• Update paddle_build.bat to remove op grad when build windows inference lib (#33800)

  Show description for 1cfa105
  OliverLPH

  authored
  1cfa105

  Copy full SHA for 1cfa105

• Enhance error message when x or y is empty in elementwise_op (#33928)

  Show description for 70100e4
  zhiqiu

  authored
  70100e4

  Copy full SHA for 70100e4

• 【HeterPS】fix hdfs and fleet_util for supporting save/load/infer (#33903)

  Show description for 2ef6188
  danleifeng

  authored
  2ef6188

  Copy full SHA for 2ef6188

• [HybridParallel] Add amp support for pipeline_parallel (#33951)

  Show description for 0b91133
  ForFishes

  authored
  0b91133

  Copy full SHA for 0b91133

• Refine the dygraph ptq and the module of calculating KL threshold (#33898)

  Show description for 9254183
  juncaipeng

  authored
  9254183

  Copy full SHA for 9254183

• fix bug of sync_parameters (#33955)

  ForFishes

  authored
  bd559a2

  Copy full SHA for bd559a2

• optimize grad add device (#33946)

  wangxicoding

  authored
  75d247b

  Copy full SHA for 75d247b

• [NPU] add abs and uniform_random op and npu dockerfile, test=develop (#33942)

  qili93

  authored
  a84e48b

  Copy full SHA for a84e48b

• [NPU] change Add to AddN in sum npu op (#33957)

  Show description for fa5ddfd
  pangyoki

  authored
  fa5ddfd

  Copy full SHA for fa5ddfd

• Add fused elemwise gelu and optimize performance (#33480)

  wangxicoding

  authored
  eae3185

  Copy full SHA for eae3185

• replace spatial with per_activation mode for bn op to improve perf (#33887)

  limin2021

  authored
  3629bf4

  Copy full SHA for 3629bf4

• make stop_gradient=True for random op in static graph (#33959)

  zhiqiu

  authored
  43876e8

  Copy full SHA for 43876e8

• [hybrid performance] optimize pipeline performance

  wangxicoding

  authored
  9914dff

  Copy full SHA for 9914dff

• add reduce_sum op into amp black list (#33960)

  Show description for aa9fdd0
  thisjiang

  authored
  aa9fdd0

  Copy full SHA for aa9fdd0

• [Dy2Stat]Fix unique_name in create_static_variable_gast_node (#33963)

  Aurelius84

  authored
  740f4e3

  Copy full SHA for 740f4e3

• [pass_enhance] : seq_concat_fc_fuse_pass (#33961)

  Wangzheee

  authored
  72af57b

  Copy full SHA for 72af57b

• correct define (#33966)

  Show description for 70ecf3b
  b3602sss

  authored
  70ecf3b

  Copy full SHA for 70ecf3b

• Reduce build time by deleting the template param BlockDim (#33901)

  ZzSean

  authored
  7a47660

  Copy full SHA for 7a47660

Commits on Jul 6, 2021

• [OP] fix histogram op when input tensor is empty, test=develop (#33970)

  qili93

  authored
  389f8c5

  Copy full SHA for 389f8c5

• Optimize the forward of log_softmax for the case when axis is not the last dimention. (#32396)

  AshburnLee

  authored
  69ffb38

  Copy full SHA for 69ffb38

• 【HETERPS】pipeline adaptive for heterps (#33159)

  Show description for bfef7fe
  danleifeng

  authored
  bfef7fe

  Copy full SHA for bfef7fe

• Enhance error message for interpolate_v2 (#33941)

  Show description for f2068ee
  tink2123

  authored
  f2068ee

  Copy full SHA for f2068ee

• [pass_enhance] embedding_eltwise_layernorm_fuse_pass (#33973)

  Wangzheee

  authored
  ae74c40

  Copy full SHA for ae74c40

• [pass_enhance] conv_elementwise_add_mkldnn_fuse_pass (#33931)

  Wangzheee

  authored
  dd33d28

  Copy full SHA for dd33d28