Commits

Branch selector

 main

User selector

reedwm

Datepicker

All time

Commit History

Commits on Aug 20, 2019

• Support non-float32 losses in LossScaleOptimizer.

  Show description for 6fb672d
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  6fb672d

  Copy full SHA for 6fb672d

• Do not cast inputs to AddLoss layers

  Show description for bb5eec0
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  bb5eec0

  Copy full SHA for bb5eec0

Commits on Aug 13, 2019

• Add mixed_float16 and mixed_bfloat16 dtype policies.

  Show description for a2a84a8
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  a2a84a8

  Copy full SHA for a2a84a8

• Add loss_scale field to Policy.

  Show description for 8cd1d56
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  8cd1d56

  Copy full SHA for 8cd1d56

Commits on Aug 9, 2019

• Disable two bfloat16 tests on Windows.

  Show description for 6e3a6bb
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  6e3a6bb

  Copy full SHA for 6e3a6bb

• Implement __repr__ for LossScale subclasses.

  Show description for d24f662
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  d24f662

  Copy full SHA for d24f662

• Improve Layer docstrings in regards to autocasting.

  Show description for 26220ec
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  26220ec

  Copy full SHA for 26220ec

Commits on Aug 5, 2019

• Do not cast arbitrary CompositeTensor inputs to a layer.

  Show description for 8a8db8e
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  8a8db8e

  Copy full SHA for 8a8db8e

Commits on Aug 3, 2019

• Disallow mixed precision to be used with TPUStrategy.

  Show description for 6b4b4d4
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  6b4b4d4

  Copy full SHA for 6b4b4d4

• Ensure AutoCastVariable.dtype is not a string.

  Show description for 063852d
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  063852d

  Copy full SHA for 063852d

Commits on Aug 2, 2019

• Fix bug where TensorFlowOpLayer would autocast inputs.

  Show description for 686c123
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  686c123

  Copy full SHA for 686c123

• Have layers autocast composite tensors.

  Show description for bd687e8
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  bd687e8

  Copy full SHA for bd687e8

• Have RNN classes pass their dtypes to their cells.

  Show description for d90e521
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  d90e521

  Copy full SHA for d90e521

• In TF 2, autocast layer inputs to layer dtype.

  Show description for e17b7ca
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  e17b7ca

  Copy full SHA for e17b7ca

Commits on Jul 31, 2019

• Add new dtype behavior for keras layers.

  Show description for b4245c3
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  b4245c3

  Copy full SHA for b4245c3

Commits on Jul 23, 2019

• Remove old incompatible versions of FusedBatchNorm ops from ops_history.*.pbtxt.

  Show description for 5c7b8ea
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  5c7b8ea

  Copy full SHA for 5c7b8ea

Commits on Jul 17, 2019

• Add mixed precision tests for Sequential/subclassed models.

  Show description for 054aabf
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  054aabf

  Copy full SHA for 054aabf

Commits on Jul 16, 2019

• Add better error message for LossScaleOptimizer AttributeError.

  Show description for c7504db
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  c7504db

  Copy full SHA for c7504db

Commits on Jul 15, 2019

• Add test where mixed precision is enabled when saving and disabled when loading, and vice versa.

  Show description for d158fbb
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  d158fbb

  Copy full SHA for d158fbb

Commits on Jul 11, 2019

• Fix issue where AutoCastVariables wouldn't cast in tf.functions.

  Show description for 53f95a9
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  53f95a9

  Copy full SHA for 53f95a9

Commits on Jul 10, 2019

• Implement AutoCastVariable's __repr__ method.

  Show description for d904d31
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  d904d31

  Copy full SHA for d904d31

• Automated rollback of commit 1ec90f6

  Show description for e737182
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  e737182

  Copy full SHA for e737182

Commits on Jul 4, 2019

• Add operator overloads to AutoCastVariable.

  Show description for 604988b
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  604988b

  Copy full SHA for 604988b

Commits on Jul 2, 2019

• Fix issue where AutoCastVariables wouldn't cast in tf.functions.

  Show description for 1ec90f6
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  1ec90f6

  Copy full SHA for 1ec90f6

• LossScaleOptimizer: Support optimizers that override apply_gradients().

  Show description for 9aa1e9b
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  9aa1e9b

  Copy full SHA for 9aa1e9b

Commits on Jun 28, 2019

• Remove old incompatible versions of FusedBatchNormGradV2 from ops_history.*.pbtxt.

  Show description for 4c7e2ed
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  4c7e2ed

  Copy full SHA for 4c7e2ed

Commits on Jun 27, 2019

• Increase shard_count for extract_image_patches_grad_test.

  Show description for b764c5d
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  b764c5d

  Copy full SHA for b764c5d

Commits on Jun 12, 2019

• Fix fp16 tf.linalg.band_part bug.

  Show description for b8b7f91
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  b8b7f91

  Copy full SHA for b8b7f91

Commits on Jun 10, 2019

• Add methods to LossScaleOptimizer to manually do loss scaling.

  Show description for 9912d06
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  9912d06

  Copy full SHA for 9912d06

• Fix error when checkpointing Layer with "infer_float32_vars".

  Show description for 04b9c11
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  04b9c11

  Copy full SHA for 04b9c11

Commits on Jun 6, 2019

• Improve some CPU Conv2D error messages.

  Show description for 6ae5b6f
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  6ae5b6f

  Copy full SHA for 6ae5b6f

• Better emulate interface of Optimizer in LossScaleOptimizer.

  Show description for 951b95d
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  951b95d

  Copy full SHA for 951b95d

Commits on Jun 4, 2019

• Fix typo in docstring

  Show description for 3b167c4
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  3b167c4

  Copy full SHA for 3b167c4

Commits on Jun 1, 2019

• Support run_eagerly=True in Keras Models with a LossScaleOptimizer

  Show description for 1d6c07b
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  1d6c07b

  Copy full SHA for 1d6c07b

Commits on May 25, 2019

• Unexpose enable_mixed_precision_graph_rewrite in TF 2.

  Show description for e79459a
  

  reedwm

  authored and

  tensorflower-gardener

  committed
  e79459a

  Copy full SHA for e79459a

Pagination

Previous

Next