Commits

Branch selector

 master

User selector

protoget

Datepicker

All time

Commit History

Commits on Jun 19, 2018

• Update a few documentation for layer-input-casting feature.

  Show description for 707ac11
  

  protoget

  authored and

  tensorflower-gardener

  committed
  707ac11

  Copy full SHA for 707ac11

Commits on Jun 14, 2018

• Remove hardcoded dtype in tf.layers.xxx() function call to make them compatible with mixed precision training apis.

  Show description for 0b8c580
  

  protoget

  authored and

  tensorflower-gardener

  committed
  0b8c580

  Copy full SHA for 0b8c580

Commits on Jun 11, 2018

• Improve the loss_scale_optimizer docstring.

  Show description for 3a1d8bd
  

  protoget

  authored and

  tensorflower-gardener

  committed
  3a1d8bd

  Copy full SHA for 3a1d8bd

Commits on Jun 8, 2018

• Replace add_variable() with add_weight() in official keras layers.

  Show description for a58cdd2
  

  protoget

  authored and

  tensorflower-gardener

  committed
  a58cdd2

  Copy full SHA for a58cdd2

Commits on May 30, 2018

• Remove unnecessary shape registration fn from cudnn rnn ops.

  Show description for 02ba495
  

  protoget

  authored and

  tensorflower-gardener

  committed
  02ba495

  Copy full SHA for 02ba495

Commits on May 17, 2018

• Add more logging in BaseGPUDevice::ComputeHelper for kernel completion.

  Show description for deca317
  

  protoget

  authored and

  tensorflower-gardener

  committed
  deca317

  Copy full SHA for deca317

• Append device name in executor logging

  Show description for b33d100
  

  protoget

  authored and

  tensorflower-gardener

  committed
  b33d100

  Copy full SHA for b33d100

Commits on May 15, 2018

• Introduce LossScalingOptimizer for mixed precision training.

  Show description for 4c2cb71
  

  protoget

  authored and

  tensorflower-gardener

  committed
  4c2cb71

  Copy full SHA for 4c2cb71

Commits on May 14, 2018

• Fix functional.While(), functional.For(rewrite_with_while)

  Show description for 1761e1d
  

  protoget

  authored and

  tensorflower-gardener

  committed
  1761e1d

  Copy full SHA for 1761e1d

Commits on May 4, 2018

• Fix build failure for macos py3

  Show description for 4d0388d
  

  protoget

  authored and

  tensorflower-gardener

  committed
  4d0388d

  Copy full SHA for 4d0388d

Commits on Apr 26, 2018

• Cudnn RNN v2 kernels with autotune capability

  Show description for 270a6e9
  

  protoget

  authored and

  tensorflower-gardener

  committed
  270a6e9

  Copy full SHA for 270a6e9

Commits on Apr 20, 2018

• Temporarily set cudnn Rnn math precision to fp32.

  Show description for 5fbb1fe
  

  protoget

  authored and

  tensorflower-gardener

  committed
  5fbb1fe

  Copy full SHA for 5fbb1fe

Commits on Apr 17, 2018

• Increase softmax gpu unittest numeric stability

  Show description for ba25b8b
  protoget

  committed
  ba25b8b

  Copy full SHA for ba25b8b

Commits on Apr 16, 2018

• Increase softmax gpu unittest numeric stability

  Show description for 90ee831
  

  protoget

  authored and

  tensorflower-gardener

  committed
  90ee831

  Copy full SHA for 90ee831

Commits on Apr 14, 2018

• Revamp Cudnn RNN kernels for incoming autotune changes.

  Show description for 360c5a3
  

  protoget

  authored and

  tensorflower-gardener

  committed
  360c5a3

  Copy full SHA for 360c5a3

Commits on Apr 13, 2018

• Add more logging for failure cases in CUDATimer

  Show description for e42ebc5
  

  protoget

  authored and

  tensorflower-gardener

  committed
  e42ebc5

  Copy full SHA for e42ebc5

• Print error msg in CUDATimer.Init() when CreateEvent() is not ok().

  Show description for 457e8b3
  

  protoget

  authored and

  tensorflower-gardener

  committed
  457e8b3

  Copy full SHA for 457e8b3

Commits on Apr 7, 2018

• Revamp a few ScratchAllocator classes in cudnn_rnn_ops

  Show description for 22cb902
  

  protoget

  authored and

  tensorflower-gardener

  committed
  22cb902

  Copy full SHA for 22cb902

Commits on Apr 6, 2018

• Support RNN profiling in StreamExecutor for CUDA GPUs.

  Show description for 4f7943f
  

  protoget

  authored and

  tensorflower-gardener

  committed
  4f7943f

  Copy full SHA for 4f7943f

Commits on Mar 21, 2018

• Change Softmax on CUDA to use fp32 for denominator when input/output are fp16.

  Show description for 942a32b
  

  protoget

  authored and

  tensorflower-gardener

  committed
  942a32b

  Copy full SHA for 942a32b

Commits on Feb 26, 2018

• Merge pull request #17222 from kbsriram/pool-gradients

  Show description for 2f05ffd
  protoget

  authored
  2f05ffd

  Copy full SHA for 2f05ffd

Commits on Feb 7, 2018

• Address Adagrad/RMSProp incompatibility with CudnnRNN

  Show description for 0282f08
  

  protoget

  authored and

  tensorflower-gardener

  committed
  0282f08

  Copy full SHA for 0282f08

Commits on Jan 5, 2018

• Automated g4 rollback of changelist 180848930

  Show description for cc94f38
  

  protoget

  authored and

  tensorflower-gardener

  committed
  cc94f38

  Copy full SHA for cc94f38

Commits on Dec 24, 2017

• Assign subgraph created by cudnn_rnn saveable to the 1st GPU.

  Show description for 56da08f
  

  protoget

  authored and

  tensorflower-gardener

  committed
  56da08f

  Copy full SHA for 56da08f

Commits on Nov 29, 2017

• Register fp16 Reduce min on GPU.

  Show description for bc87c28
  

  protoget

  authored and

  tensorflower-gardener

  committed
  bc87c28

  Copy full SHA for bc87c28

Commits on Nov 14, 2017

• Delete infeasible TODO in CudnnRNN

  Show description for 579276a
  

  protoget

  authored and

  tensorflower-gardener

  committed
  579276a

  Copy full SHA for 579276a

Commits on Nov 13, 2017

• Fix CuDNNCompatibleGRU after GRUCell refactorization

  Show description for 90222dd
  

  protoget

  authored and

  tensorflower-gardener

  committed
  90222dd

  Copy full SHA for 90222dd

• Fix build breakage

  Show description for 4d5e805
  

  protoget

  authored and

  tensorflower-gardener

  committed
  4d5e805

  Copy full SHA for 4d5e805

Commits on Nov 11, 2017

• Include memory size info in error message.

  Show description for 8d15389
  

  protoget

  authored and

  aselle

  committed
  8d15389

  Copy full SHA for 8d15389

Commits on Nov 3, 2017

• Switch tf.contrib.cudnn_rnn.CudnnXXX to point to layer APIs instead of op wrappers

  Show description for 823d8d4
  

  protoget

  authored and

  tensorflower-gardener

  committed
  823d8d4

  Copy full SHA for 823d8d4

• Support Cudnn RNN Fp16

  Show description for 509d0f2
  

  protoget

  authored and

  tensorflower-gardener

  committed
  509d0f2

  Copy full SHA for 509d0f2

Commits on Oct 26, 2017

• Switch tf.contrib.cudnn_rnn.CudnnXXX to point to layer APIs instead of op wrappers

  Show description for 62f62bf
  

  protoget

  authored and

  tensorflower-gardener

  committed
  62f62bf

  Copy full SHA for 62f62bf

Commits on Oct 6, 2017

• Fix a minor issue w/ allreduce

  Show description for af6e00f
  

  protoget

  authored and

  tensorflower-gardener

  committed
  af6e00f

  Copy full SHA for af6e00f

• Fix a minor issue w/ allreduce

  Show description for 71a2859
  

  protoget

  authored and

  tensorflower-gardener

  committed
  71a2859

  Copy full SHA for 71a2859

Commits on Oct 5, 2017

• Introduce CudnnRNN layers

  Show description for 2ae5bfc
  

  protoget

  authored and

  tensorflower-gardener

  committed
  2ae5bfc

  Copy full SHA for 2ae5bfc

Pagination

Previous

Next