- 26 3月, 2020 2 次提交
-
-
由 Reed Wanderman-Milne 提交于
I first submitted this in 3931d393 but was rolled back since Nones were filtered out from the gradients, but not the variables. I now add Nones back to the gradients so they properly match up. PiperOrigin-RevId: 302107549 Change-Id: I81b7fb71c9cdaa458475d83f784366ce8405fb74
-
由 Ran Chen 提交于
CentralStorageStrategy PiperOrigin-RevId: 302804311 Change-Id: Ibb27c529251390f40338cd296537cd98f8940b56
-
- 13 3月, 2020 1 次提交
-
-
由 Ken Franko 提交于
- experimental_run_v2 -> run PiperOrigin-RevId: 300574367 Change-Id: I5d82ea5450a4d32aea6d05ed3db4f02b8edb2eea
-
- 06 3月, 2020 1 次提交
-
-
由 Thomas O'Malley 提交于
PiperOrigin-RevId: 299150128 Change-Id: Ie0ef99dcbd1afc91ad1a0e19c56638c5e48a7865
-
- 28 2月, 2020 1 次提交
-
-
由 Reed Wanderman-Milne 提交于
The experimental_run_tf_function parameter no longer has any effect. I didn't remove the functionality from testing_util.py and keras_parameterized.py to run with experimental_run_tf_function being True and False. I will remove that functionality in a future change. PiperOrigin-RevId: 297674422 Change-Id: I5b1e67f78b4c3b60242241fb4dc2018f0ace6013
-
- 25 2月, 2020 1 次提交
-
-
由 Rick Chao 提交于
PiperOrigin-RevId: 297021321 Change-Id: I250c02ccdab46892285abe9fd7a86d9034faa1c3
-
- 22 2月, 2020 2 次提交
-
-
由 A. Unique TensorFlower 提交于
Makes Model compile run inside the captured distribution strategy, so that any metrics/optimizers created by compile are created in the distribution strategy scope (e.g. when deserializing strings that are the metric names). Also adds a correctness test that verifies the model successfully captures the distribution strategy. It also raises an error if there are metrics that are created outside the scope and not in compile. So e.g. this will help the following case because the optimizer/metrics get created by compile: with strategy.scope(): model = ... model.compile(optimizer='sgd', metrics=['binary_accuracy']) And, it will raise an error in the following case because the metrics are created in a different distribution strategy scope than the model: with strategy.scope(): model = ... model.compile(optimizer=tf.keras.optimizers.Blah() metrics=[tf.keras.metrics.BinaryAccuracy()]) PiperOrigin-RevId: 296553610 Change-Id: I988a80c1863de732da45555c444fc5a237ecd425
-
由 Håkon Sandsmark 提交于
Some of the GPU strategies failed earlier with another error message.
-
- 21 2月, 2020 2 次提交
-
-
由 Håkon Sandsmark 提交于
When calling `Optimizer.apply_gradients()` in a cross-replica distribution context (with a non-default distribution strategy), `distribute_ctx.get_replica_context()` returns None, so it would fail with the error [...]/optimizer_v2.py", line 448, in apply_gradients return distribute_ctx.get_replica_context().merge_call( AttributeError: 'NoneType' object has no attribute 'merge_call' This commit changes the error to a `RuntimeError` with a more descriptive error message (inspired by the error message in the v1 optimizer) guiding the user how to fix the issue, by either calling the `_distributed_apply()` function instead or by using `tf.distribute.Strategy.experimental_run_v2`.
-
由 Rick Chao 提交于
Supplement multi_worker_callback_tf2_test.py with tests that have the same file path for ModelCheckpoint and TensorBoard callbacks. PiperOrigin-RevId: 296328002 Change-Id: I28bb1d4b60e1fb47c1570852fe82d71adc6ebffe
-
- 20 2月, 2020 1 次提交
-
-
由 Thomas O'Malley 提交于
Kept all new abstractions private for now. In a few weeks, if we're comfortable that these abstractions are working and stable, we should expose many of them publicly. Capabilites added by this CL: (1) Easy to create a custom training step via overriding Model._train_step (2) Easy to create custom tf.function / DistStrat logic via overriding Model._make_train_function (3) Advanced users can override Model.compile and Model.fit (4) Full support for dicts, nested structures, etc with Subclassed Models. (5) "Power user" path (tf.data inputs) only modifies data in Model._train_step, where this behavior is easy to override and disable. This applies even to Keras's assumption that data is passed in (x, y, sample_weight) format. Behavior changes: (1) "loss" passed to Callbacks is now stateful (like all other metrics in Callbacks). This greatly simplifies the training step logic and callback logic. (2) ProgbarLogger always uses steps. If steps is not available, the ProgbarLogger handles inferring the steps after the first epoch. (3) validation_batch_size added in `fit`, rather than inferring from generator. (4) Model.inputs, Model.outputs, Model.input_names, and Model.output_names are no longer populated for subclassed Models. Instead, "pseudo" output names are created for subclassed Models, which are only used for metrics names and SavedModel's signature. (5) Cast NumPy floats to backend.floatx(), otherwise leave unchanged (this is likely not a change, we did something like this in our old version but the logic was scattered in many places) PiperOrigin-RevId: 296090972 Change-Id: Ia5ac833fd39085bddb016833bd338083d0dc5fc2
-
- 15 2月, 2020 1 次提交
-
-
由 Scott Zhu 提交于
PiperOrigin-RevId: 295171678 Change-Id: Ia93dd94ac2f28bb2bdc1e346961e1d138e3ba4db
-
- 14 2月, 2020 1 次提交
-
-
由 Gunhan Gulsoy 提交于
PiperOrigin-RevId: 295072178 Change-Id: Ic2454e70e12e3bea76a2094273144e6b5ae57ffb
-
- 13 2月, 2020 1 次提交
-
-
由 Rick Chao 提交于
Fix multi_worker_callback_tf2_test test target by only running it with CPU. The test is not using GPU anyway. PiperOrigin-RevId: 294784694 Change-Id: I9e9d2f8db05160799ef64b43f0cc8b1a927637e0
-
- 12 2月, 2020 2 次提交
-
-
由 A. Unique TensorFlower 提交于
It was difficult to change internal op counts because there are many tests that depend on the specific seed that's currently based on op counts. With this change, get_seed() doesn't depend on the internal op count but the number of calls to get_seed() PiperOrigin-RevId: 294523232 Change-Id: I3dc05a8aed6d42dcc372b734615312eb94aea81d
-
由 Ken Franko 提交于
PiperOrigin-RevId: 294472673 Change-Id: Ifbd6d61aa31dbb1a692654ac3c0491c2be6253bc
-
- 11 2月, 2020 1 次提交
-
-
由 Kazuaki Ishizaki 提交于
-
- 08 2月, 2020 1 次提交
-
-
由 Ken Franko 提交于
PiperOrigin-RevId: 293916796 Change-Id: Ic76ca27fb1c11c6f612eaf72ac1199e6499df5a4
-
- 04 2月, 2020 2 次提交
-
-
由 Rick Chao 提交于
PiperOrigin-RevId: 293024458 Change-Id: I639f3cf4cd8d897fc26e7ec70aa8ed3246c9de95
-
由 Gunhan Gulsoy 提交于
PiperOrigin-RevId: 292974875 Change-Id: I4067f019c8ba63090a971fd7d32840d883ab34f6
-
- 02 2月, 2020 1 次提交
-
-
由 Anjali Sridhar 提交于
Add support for aggregating batch statistics across devices by using the newly added tf.keras.layers.experimental.SyncBatchNormalization layer. PiperOrigin-RevId: 292723222 Change-Id: I1c0458ec24c7e712ffa5e12dcf1f5efd6b4ce8ac
-
- 01 2月, 2020 1 次提交
-
-
由 Yanhui Liang 提交于
PiperOrigin-RevId: 292646148 Change-Id: Ieaa51721c20fd56b9980bdedc95755c8687daa96
-
- 25 1月, 2020 1 次提交
-
-
由 Reed Wanderman-Milne 提交于
PiperOrigin-RevId: 291514265 Change-Id: I0540d980fdf0cb6222273ff48f6f72223ba90ff2
-
- 18 1月, 2020 1 次提交
-
-
由 Yanhui Liang 提交于
PiperOrigin-RevId: 290397331 Change-Id: Id88083fff97a0bed45829952d10812cd0ddfa91b
-
- 16 1月, 2020 1 次提交
-
-
由 Anjali Sridhar 提交于
PiperOrigin-RevId: 289922514 Change-Id: I2f82cade789d707f287b9915d9856e2683aaa9f6
-
- 10 1月, 2020 2 次提交
- 21 12月, 2019 1 次提交
-
-
由 Zhenyu Tan 提交于
PiperOrigin-RevId: 286639024 Change-Id: Ib28435a59bd54f8916a9b8055b7a562eb1f2f2bc
-
- 20 12月, 2019 1 次提交
-
-
由 Deven Desai 提交于
-
- 19 12月, 2019 1 次提交
-
-
由 Brian Atkinson 提交于
This is mostly the result of an internal cleanup and formatting pass. PiperOrigin-RevId: 286318018 Change-Id: I8f9e2f7519070035da73f9f24d2fc90864abc51b
-
- 17 12月, 2019 1 次提交
-
-
由 Chris Jones 提交于
PiperOrigin-RevId: 285974579 Change-Id: Ib83e53d6e34a7def18f5ac55c4d2838ca91b799e
-
- 13 12月, 2019 2 次提交
-
-
由 Rick Chao 提交于
PiperOrigin-RevId: 285287989 Change-Id: I7f4c66c344de33d16c4d572be50aed90a4d8faaf
-
由 Brian Atkinson 提交于
PiperOrigin-RevId: 285283853 Change-Id: I2534d9fb51955cc9a86d1900ec60fc265f451ddc
-
- 10 12月, 2019 1 次提交
-
-
由 Brian Zhao 提交于
This is part of the refactoring described in the Tensorflow Build Improvements RFC: https://github.com/tensorflow/community/pull/179 Subsequent changes will migrate targets from build_refactor.bzl into the new BUILD files. PiperOrigin-RevId: 284712709 Change-Id: I650eb200ba0ea87e95b15263bad53b0243732ef5
-
- 06 12月, 2019 1 次提交
-
-
由 Thomas O'Malley 提交于
PiperOrigin-RevId: 284078361 Change-Id: Iaa163487b4acee59077b5bb4c2232f5742d0e9ba
-
- 02 12月, 2019 2 次提交
-
-
由 A. Unique TensorFlower 提交于
PiperOrigin-RevId: 283310504 Change-Id: Ie7af08bdc52c660d61ab527d72f616c088ad1480
-
由 Chris Jones 提交于
PiperOrigin-RevId: 283308832 Change-Id: I7d1c4fd981a29fa07f45317539cc9e4ef120c308
-
- 23 11月, 2019 1 次提交
-
-
由 Rachel Lim 提交于
PiperOrigin-RevId: 282063336 Change-Id: If1a71cc5721be130d6fbe4bfde06008158729db9
-
- 19 11月, 2019 1 次提交
-
-
由 Rick Chao 提交于
Remove usage of multi_process_runner_util.try_run_and_except_connection_error() as setting FAIL_FAST=false fixes the connection error issue. Make grpc_fail_fast an arg for MultiProcessRunner's __init__(). PiperOrigin-RevId: 281193957 Change-Id: I1c99cb90a15fdb26892d0ad37c533c186297b09d
-
- 12 11月, 2019 1 次提交
-
-
由 Rick Chao 提交于
PiperOrigin-RevId: 279830188 Change-Id: I19d7f2f1a610817522563248d0948611f3f49328
-