- 16 10月, 2018 23 次提交
-
-
由 Chris Jones 提交于
Fixes a bug in tf.train.Saver(), where classes using the `VARIABLE_VALUE_KEY` used different naming in the checkpoint file when `var_list` was a dict. PiperOrigin-RevId: 217182136
-
由 TensorFlower Gardener 提交于
PiperOrigin-RevId: 217180014
-
由 TensorFlower Gardener 提交于
PiperOrigin-RevId: 217179947
-
由 TensorFlower Gardener 提交于
PiperOrigin-RevId: 217179854
-
由 TensorFlower Gardener 提交于
PiperOrigin-RevId: 217176593
-
由 Peter Hawkins 提交于
PiperOrigin-RevId: 217173355
-
由 Jian Li 提交于
PiperOrigin-RevId: 217171519
-
由 Martin Wicke 提交于
PiperOrigin-RevId: 217170873
-
由 Peter Hawkins 提交于
PiperOrigin-RevId: 217170781
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 217168657
-
由 Peter Hawkins 提交于
PiperOrigin-RevId: 217168404
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 217166408
-
由 Michael Kuperstein 提交于
PiperOrigin-RevId: 217164296
-
由 Eugene Brevdo 提交于
PiperOrigin-RevId: 217163302
-
由 Dan Moldovan 提交于
PiperOrigin-RevId: 217161539
-
由 Russell Power 提交于
PiperOrigin-RevId: 217159671
-
由 Ben Zinberg 提交于
PiperOrigin-RevId: 217157127
-
由 TensorFlower Gardener 提交于
PiperOrigin-RevId: 217155008
-
由 Allen Lavoie 提交于
Correctly handles things that pretend to be lists, like the ListWrappers we use for variable tracking. Fixes #22853. PiperOrigin-RevId: 217154951
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 217153168
-
由 TensorFlower Gardener 提交于
PiperOrigin-RevId: 217152548
-
由 Adrian Kuegel 提交于
PiperOrigin-RevId: 217151877
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 217151569
-
- 15 10月, 2018 7 次提交
-
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 217148772
-
由 Gunhan Gulsoy 提交于
PiperOrigin-RevId: 217147663
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 217147034
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 217145701
-
由 Mihai Maruseac 提交于
PiperOrigin-RevId: 217144630
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 217108509
-
由 A. Unique TensorFlower 提交于
[TF CTC] Fix Returns documentation stanzas to be properly formatted (see https://www.tensorflow.org/api_docs/python/tf/nn/ctc_greedy_decoder) PiperOrigin-RevId: 217100215
-
- 14 10月, 2018 8 次提交
-
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 217034982
-
由 Gunhan Gulsoy 提交于
This makes the script more platform independent. PiperOrigin-RevId: 217028598
-
由 Tom Hennigan 提交于
PiperOrigin-RevId: 217025180
-
由 Yu-Cheng Ling 提交于
PiperOrigin-RevId: 217020295
-
由 A. Unique TensorFlower 提交于
Disables logging and summaries of steps, steps/s, loss etc. when RunConfig.log_step_count_steps is None. This is already how Estimator works (as one can see in line 2136 where we explicitly replace log_step_count_steps as None). It should be backward compatible as current code would have crashed if log_step_count_steps is None. PiperOrigin-RevId: 217019985
-
由 Rohan Jain 提交于
The underlying issue with NaN's has now been resolved. PiperOrigin-RevId: 217014692
-
由 A. Unique TensorFlower 提交于
In TPUEstimator, allow input_fn to return features and labels of any arbitrary nested structure rather than just tensor or dict or tensors. The current restriction seems to be historical. The code in _ModelFnWrapper uses nest flattens already, and then in _call_model_fn, 'features' and 'labels' are passed directly to the user code model_fn, so any nested structure should be supported. Before entering ModelFnWrapper, the _InputPipeline and _Inputs classes also both handle nested structure as well. PiperOrigin-RevId: 217010800
-
由 A. Unique TensorFlower 提交于
This requires #ifdef'ing out the code to use JSON parsing/serialization. PiperOrigin-RevId: 217003132
-
- 13 10月, 2018 2 次提交
-
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 216976797
-
由 Benjamin Kramer 提交于
The tuple buffer is never read, so stop emitting code to fill it. A typical root tuple consists of a H2D memcpy and a host callback, both of which are somewhat slow. This helps tiny models and inference benchmarks, where the host/device syncs can be a significant part of the runtime of the entire computation. PiperOrigin-RevId: 216968475
-