Skip to content

T
Tensorflow

wux_labs / Tensorflow

通知 1
Star 0
Fork 0
T
Tensorflow

体验新版 GitCode，发现更多精彩内容 >>

切换分支/标签
  1. 10 8月, 2017 27 次提交
  3. 09 8月, 2017 13 次提交
    • A
      Relax tolerance to fix OSS test failure on MacOS. · e3034efc
      由 A. Unique TensorFlower 提交于 
      PiperOrigin-RevId: 164728247
      e3034efc
    • H
      Consider the nested computations when checking if an instruction is · b4001ea6
      由 HyoukJoong Lee 提交于 
      removable from a computation. This is to prevent DCE from removing a
while instruction that includes a send/recv instruction.

PiperOrigin-RevId: 164722478
      b4001ea6
    • A
      Remove obsolete advice on BUILD flags · acceb27d
      由 A. Unique TensorFlower 提交于 
      PiperOrigin-RevId: 164718342
      acceb27d
    • J
      slot_creator: fix bugs handling dynamic-shaped var/tensor · e6e7ee49
      由 James Qin 提交于 
      PiperOrigin-RevId: 164686075
      e6e7ee49
    • R
      Remove newlines from tf.nn.dynamic_rnn args list. · 26e628b8
      由 RJ Ryan 提交于 
      Prevents bad formatting: https://www.tensorflow.org/versions/r1.2/api_docs/python/tf/nn/dynamic_rnn

PiperOrigin-RevId: 164675585
      26e628b8
    • B
      Use hand crafted filter instead of regexp since regexp don't always work · bb23f540
      由 Benoit Steiner 提交于 
      properly on some platforms.

PiperOrigin-RevId: 164665656
      bb23f540
    • Y
      Infer shapes for RestoreV2 and RestoreSlice ops when shape_and_slice input is present. · 8e6c372f
      由 Yuefeng Zhou 提交于 
      PiperOrigin-RevId: 164660701
      8e6c372f
    • A
      Unique names in the batch_function decorator. · 8d23b781
      由 Alexandre Passos 提交于 
      PiperOrigin-RevId: 164659904
      8d23b781
    • A
      Make plugin_data an optional field of SummaryMetadata · 4c60c962
      由 A. Unique TensorFlower 提交于 
      Every summary op writes data for a single plugin to process. Hence, each SummaryMetadata proto should have a single PluginData optional field (instead of a repeated one). This removes much complexity from TensorBoard logic that loops over the plugin data. It also simplifies the SQL schema - it can now enforce a one-to-one relationship between summary op and plugin.

PiperOrigin-RevId: 164659570
      4c60c962
    • F
      Make a change to the Cluster Resolver API: If no `credentials` are passed in... · de5034ae
      由 Frank Chen 提交于 
      Make a change to the Cluster Resolver API: If no `credentials` are passed in to the GCE and TPU Cluster Resolvers, then we will use the GoogleCredentials.get_application_default() credentials. If users want to pass in no credentials at all, then they will have to pass in "None" explicitly.

PiperOrigin-RevId: 164659129
      de5034ae
    • D
      [tf.contrib.data] Enable using step-local resources in Dataset.map()/filter(). · 865b92da
      由 Derek Murray 提交于 
      This change ensures that the mapper/predicate function used
respectively in these transformations has its own ScopedStepContainer,
thereby allowing the use of TensorArray resources (and operations that
use them, such as control-flow ops) inside these functions.

Fixes #11715.

PiperOrigin-RevId: 164648309
      865b92da
    • A
      Speed up tf.determinant by using LU factorization kernels from cuSolver for... · 389a7d43
      由 A. Unique TensorFlower 提交于 
      Speed up tf.determinant by using LU factorization kernels from cuSolver for large matrices instead of the batched LU factorization from cuBlas, which is only suitable for small matrices.

Speedup measured on Titan X (Maxwell):

Shape            Before    After    Speedup
------------------------------------------------------
(4, 4)          0.000159   0.000200 -26.35% (noise)
(16, 16)        0.000198   0.000190   3.59%
(64, 64)        0.000592   0.000538   9.10%
(128, 128)      0.001348   0.001376  -2.14%
(200, 200)      0.003201   0.002882   9.94%
(256, 256)      0.005096   0.003373  33.81%
(1024, 1024)    0.169690   0.012452  92.66%
(2, 512, 512)   0.023370   0.012243  47.61%
(2, 1024, 1024) 0.178757   0.025198  85.90%
(4, 4, 4)       0.000121   0.000128  -5.79%
(4, 16, 16)     0.000212   0.000190   9.95%
(4, 64, 64)     0.000499   0.000514  -3.01%
(4, 128, 128)   0.001276   0.001214   4.79%
(4, 256, 256)   0.004364   0.004314   1.14%
(4, 512, 512)   0.025031   0.024956   0.30%
(4, 1024, 1024) 0.184210   0.052858  71.31%
(8, 512, 512)   0.026542   0.026502   0.15%
(8, 1024, 1024) 0.186145   0.185988   0.08%
(65, 4, 4)      0.000152   0.000142   6.05%
(65, 16, 16)    0.000197   0.000194   1.52%
(65, 64, 64)    0.000559   0.000549   1.79%
(65, 128, 128)  0.001326   0.001308   1.29%
(65, 256, 256)  0.005495   0.005525  -0.55%
(65, 512, 512)  0.034147   0.034662  -1.51%
(513, 4, 4)     0.000144   0.000195 -35.42% (noise)
(513, 16, 16)   0.000207   0.000200   3.38%
(513, 64, 64)   0.001502   0.001490   0.79%
(513, 256, 256) 0.033428   0.032933   1.48%
(513, 512, 512) 0.234707   0.216858   7.60%

PiperOrigin-RevId: 164633730
      389a7d43
    • A
      Speed up GPU version of tf.matrix_inverse by using LU factorization kernels... · e57e11b7
      由 A. Unique TensorFlower 提交于 
      Speed up GPU version of tf.matrix_inverse by using LU factorization kernels from cuSolver and a hand-written matrix identity kernel, instead of the batched LU factorization from cuBlas, which is only suitable for small matrices.

Speedup measured on Titan X (Maxwell):

Shape           adjoint    Before    After    Speedup
------------------------------------------------------
(4, 4)          noadjoint  0.000204  0.000193   5.3%
(16, 16)        noadjoint  0.000360  0.000186  48.3%
(256, 256)      noadjoint  0.013830  0.003852  72.1%
(1024, 1024)    noadjoint  0.647639  0.015075  97.6%
(513, 4, 4)     noadjoint  0.000219  0.000192  12.3%
(513, 16, 16)   noadjoint  0.000293  0.000195  33.4%
(513, 256, 256) noadjoint  0.120573  0.120175   0.3%
(4, 4)          adjoint    0.000201  0.000193   3.9%
(16, 16)        adjoint    0.000282  0.000185  34.4%
(256, 256)      adjoint    0.013028  0.003391  73.9%
(1024, 1024)    adjoint    0.647752  0.014341  97.7%
(513, 4, 4)     adjoint    0.000221  0.000197  10.8%
(513, 16, 16)   adjoint    0.000384  0.000205  46.6%
(513, 256, 256) adjoint    0.131402  0.130616   0.6%

PiperOrigin-RevId: 164623298
      e57e11b7