| 6cc140e | Henning Becker | 02 April 2024, 16:38:14 UTC | Make Sonnet use CudnnRNNV3 CudnnRNN and CudnnRNNV2 are not compatible with cuDNN 9+, so this change makes Sonnet use CudnnRNNV3 instead. Note that this raises the minimum supported cuDNN version to 8.1 (which is below 8.9 - the minimum supported cuDNN version in Tensorflow anyway). PiperOrigin-RevId: 621206198 | 02 April 2024, 16:39:00 UTC |
| 26b0518 | Tom Hennigan | 02 January 2024, 11:17:13 UTC | Bump version to 2.0.3.dev. PiperOrigin-RevId: 595072254 | 02 January 2024, 11:17:47 UTC |
| 95d8158 | Tom Hennigan | 02 January 2024, 10:54:53 UTC | Cut Sonnet 2.0.2 release. PiperOrigin-RevId: 595068296 | 02 January 2024, 10:55:21 UTC |
| 03ab185 | Tom Hennigan | 02 January 2024, 10:27:52 UTC | Fix import ordering. PiperOrigin-RevId: 595063851 | 02 January 2024, 10:28:33 UTC |
| 12a7c79 | Richard Levasseur | 21 July 2023, 09:08:43 UTC | Internal change PiperOrigin-RevId: 549884461 | 21 July 2023, 09:09:17 UTC |
| ee6c95c | Fiona Lang | 30 June 2023, 17:46:08 UTC | Update direct import of TF to its correct location instead of importing directly from the python init file. This is in preparation for deleting imports from the python init file. PiperOrigin-RevId: 544696304 | 30 June 2023, 17:46:46 UTC |
| 26a685e | Sergei Lebedev | 23 February 2023, 15:38:33 UTC | Switched from the deprecated defun_with_attributes() to tf.function() PiperOrigin-RevId: 511782775 | 23 February 2023, 15:39:10 UTC |
| ee36d84 | Sergei Lebedev | 23 February 2023, 12:13:10 UTC | Do not test on 3.7 PiperOrigin-RevId: 511748287 | 23 February 2023, 12:13:46 UTC |
| 384eae3 | Sergei Lebedev | 23 February 2023, 11:44:04 UTC | Updated supported Python versions PiperOrigin-RevId: 511743612 | 23 February 2023, 11:44:35 UTC |
| bb650b9 | Tom Hennigan | 15 December 2022, 15:11:22 UTC | Bump Sonnet version to 2.0.2.dev. PiperOrigin-RevId: 495584613 | 15 December 2022, 15:11:58 UTC |
| 950ecb1 | Tom Hennigan | 15 December 2022, 14:48:59 UTC | Bump Sonnet version to 2.0.1. PiperOrigin-RevId: 495579932 | 15 December 2022, 14:49:36 UTC |
| 45c47e6 | Faizan Muhammad | 02 November 2022, 20:49:33 UTC | Remove functools.wraps for unbound methods to support inspect.signature PiperOrigin-RevId: 485685579 | 02 November 2022, 20:50:05 UTC |
| 6b90fdb | Umer Javed | 18 October 2022, 16:30:31 UTC | Removed call to function_lib's register method PiperOrigin-RevId: 481941823 | 18 October 2022, 16:31:22 UTC |
| 7298417 | Sonnet Contributor | 23 August 2022, 20:08:23 UTC | BUILD cleanup PiperOrigin-RevId: 469537143 | 23 August 2022, 20:08:58 UTC |
| c92ac8e | Chen Qian | 16 August 2022, 19:22:00 UTC | Use legacy keras optimizer to be compatible with an incoming Keras optimizer migration. PiperOrigin-RevId: 467992914 | 16 August 2022, 19:22:29 UTC |
| d1cd371 | Yilei Yang | 07 March 2022, 21:15:40 UTC | Update test for Python 3.9. PiperOrigin-RevId: 433021769 | 07 March 2022, 21:16:14 UTC |
| df5d099 | Sonnet Contributor | 01 February 2022, 18:27:44 UTC | Replace np.bool with np.bool_. As of NumPy 1.20, `np.bool` is an alias of builtin type `bool`. The numpy scalar type is `np.bool_`. PiperOrigin-RevId: 425658873 | 01 February 2022, 18:28:37 UTC |
| 5cbfdc3 | Tom Hennigan | 15 December 2021, 15:35:49 UTC | Defer creation of autograph converted functions until usage. Some users (e.g. #226) report issues with autograph. While these can't be fixed by Sonnet, we can defer the creation of autograph converted functions until they are first used (e.g. until the first time you use batch norm with `@tf.function`). This enables users to make use of non-autograph requiring features in Sonnet and may be a sufficient workaround while TF team debug. Fixes #226. PiperOrigin-RevId: 416554741 | 15 December 2021, 15:36:30 UTC |
| 958ea1b | Tom Hennigan | 13 December 2021, 12:51:03 UTC | Bump Sonnet version to `2.0.1.dev`. PiperOrigin-RevId: 416012688 | 13 December 2021, 12:51:34 UTC |
| 913cbd3 | Mehdi Amini | 30 November 2021, 10:17:47 UTC | Disable flaky test from CI PiperOrigin-RevId: 413094278 | 30 November 2021, 10:18:21 UTC |
| 19e238e | Mehdi Amini | 17 October 2021, 14:26:53 UTC | Tag flaky test sonnet/v2/src/conformance:checkpoint_test_gpu as "notap" PiperOrigin-RevId: 403759548 | 17 October 2021, 14:27:22 UTC |
| 892790c | Copybara-Service | 08 September 2021, 08:52:39 UTC | Merge pull request #214 from deepmind:dependabot/pip/tensorflow-2.5.1 PiperOrigin-RevId: 395424166 | 08 September 2021, 08:52:39 UTC |
| 0b550ff | Sergei Lebedev | 02 September 2021, 10:03:58 UTC | Added __init__.py files to all src subpackages PiperOrigin-RevId: 394419068 | 02 September 2021, 10:04:40 UTC |
| 57f4462 | dependabot[bot] | 25 August 2021, 14:45:08 UTC | Bump tensorflow from 2.5.0 to 2.5.1 Bumps [tensorflow](https://github.com/tensorflow/tensorflow) from 2.5.0 to 2.5.1. - [Release notes](https://github.com/tensorflow/tensorflow/releases) - [Changelog](https://github.com/tensorflow/tensorflow/blob/master/RELEASE.md) - [Commits](https://github.com/tensorflow/tensorflow/compare/v2.5.0...v2.5.1) --- updated-dependencies: - dependency-name: tensorflow dependency-type: direct:production ... Signed-off-by: dependabot[bot] <support@github.com> | 25 August 2021, 14:45:08 UTC |
| 0e25f47 | Sonnet Contributor | 13 August 2021, 18:11:51 UTC | N/A BUILD file cleanup PiperOrigin-RevId: 390652227 | 13 August 2021, 18:12:21 UTC |
| 38adf3a | Tom Hennigan | 12 August 2021, 15:56:21 UTC | Mark embeddings on VectorQuantizerEMA as non-trainable. This variable is updated as part of the forward pass (during training) and is not updated by the optimiser. In TF `trainable` is used to indicate variables that should be managed by an optimizer, so even though we update this during training and it is a parameter of our model, we should set `trainable=False`. Fixes #209. PiperOrigin-RevId: 390382636 | 12 August 2021, 15:56:52 UTC |
| c87468e | Sergei Lebedev | 18 June 2021, 19:01:55 UTC | Removed six Sonnet 2 does not support 2.X, so it makes little sense to keep six around. PiperOrigin-RevId: 380235488 | 18 June 2021, 19:02:30 UTC |
| cf15370 | Tom Hennigan | 18 June 2021, 12:46:27 UTC | Remove Python 2 supporting code. PiperOrigin-RevId: 380168270 | 18 June 2021, 12:46:56 UTC |
| 9bb751f | Tom Hennigan | 18 June 2021, 09:55:21 UTC | Add functional API to Sonnet 2 inspired by JAX [0] and Haiku [1]. [0] https://github.com/google/jax [1] https://github.com/deepmind/dm-haiku PiperOrigin-RevId: 380147873 | 18 June 2021, 09:55:58 UTC |
| 6c2bf5f | Tom Hennigan | 10 June 2021, 10:05:52 UTC | Add pytest github action and get tests passing. PiperOrigin-RevId: 378611382 | 10 June 2021, 10:06:20 UTC |
| 1a5a6c4 | Tom Hennigan | 09 June 2021, 11:24:02 UTC | Relax result checking for FP values. CUDA-11 produces slightly different results and causes the test failure otherwise. PiperOrigin-RevId: 378374202 | 09 June 2021, 11:24:48 UTC |
| 81baad2 | Rebecca Chen | 01 June 2021, 18:28:39 UTC | Add missing typing.Optional type annotations to function parameters. PiperOrigin-RevId: 376879768 Change-Id: I6efe84d5f989d8d5663d7ab6b5b3559fe071ade3 | 01 June 2021, 18:29:10 UTC |
| c9cdbb4 | Eugene Zhulenev | 15 March 2021, 18:22:54 UTC | Fix sonnet/v2/src/nets/dnc:util_test_gpu test PiperOrigin-RevId: 362986883 Change-Id: If791985603e97af2f0bca8c2255a2ba995f27b28 | 15 March 2021, 18:23:31 UTC |
| 01ef9a6 | Yilei Yang | 11 March 2021, 15:29:15 UTC | Remove no-op pylint disable comments. PiperOrigin-RevId: 362287115 Change-Id: Ib5efc634f54914055d10b914f7aac70d386edf7e | 11 March 2021, 15:29:48 UTC |
| 76f61d5 | Tamara Norman | 11 March 2021, 14:35:47 UTC | Remove tensorflow-gpu from extra requirements Also update dependencies in requirements-tf.txt PiperOrigin-RevId: 362278786 Change-Id: Ib8943e04c36a7fe31782aecaede8cd98bfd69347 | 11 March 2021, 14:36:11 UTC |
| eb06599 | Isha Arkatkar | 10 February 2021, 05:14:47 UTC | Fix the checkpoint_test after fix for bug b/168905859 PiperOrigin-RevId: 356659488 Change-Id: Ibd52064ff9c343cf10eee5bb2720e04e10ccb421 | 10 February 2021, 05:15:36 UTC |
| 5a46569 | Copybara-Service | 08 October 2020, 10:01:11 UTC | Merge pull request #181 from tirkarthi:fix-collections PiperOrigin-RevId: 336048465 Change-Id: I93e579d4e9832aa9ca66a68fac04ea88183970ee | 08 October 2020, 10:01:11 UTC |
| af46202 | Copybara-Service | 01 October 2020, 13:24:02 UTC | Merge pull request #190 from deepmind:dependabot/pip/tensorflow-2.2.1 PiperOrigin-RevId: 334799980 Change-Id: Ie453fa1ca5cf42d06c5d1bfdd952952cee518011 | 01 October 2020, 13:24:02 UTC |
| c3e07d3 | Sonnet Contributor | 30 September 2020, 22:58:35 UTC | BUILD cleanup PiperOrigin-RevId: 334697535 Change-Id: I738c37f5b79f02bd9ab5d0c027858293ffbeb58a | 01 October 2020, 13:23:41 UTC |
| 91a4c09 | Lena Martens | 14 September 2020, 13:55:50 UTC | Internal change. PiperOrigin-RevId: 331537710 Change-Id: Ie0374ee84cc781870ad0717b59a6aaacbfa3fc44 | 01 October 2020, 13:23:23 UTC |
| af3826a | dependabot[bot] | 01 October 2020, 10:15:16 UTC | Bump tensorflow from 2.2.0 to 2.2.1 Bumps [tensorflow](https://github.com/tensorflow/tensorflow) from 2.2.0 to 2.2.1. - [Release notes](https://github.com/tensorflow/tensorflow/releases) - [Changelog](https://github.com/tensorflow/tensorflow/blob/master/RELEASE.md) - [Commits](https://github.com/tensorflow/tensorflow/compare/v2.2.0...v2.2.1) Signed-off-by: dependabot[bot] <support@github.com> | 01 October 2020, 10:15:16 UTC |
| 4d16e10 | Sonnet Contributor | 09 September 2020, 12:35:53 UTC | Internal refactor PiperOrigin-RevId: 330704150 Change-Id: Ib372f470428f41b2656a6d680d86bca73b973a30 | 01 October 2020, 09:20:16 UTC |
| 3241e36 | Sonnet Contributor | 07 September 2020, 18:51:40 UTC | Internal change PiperOrigin-RevId: 330403382 Change-Id: Ie0ea28603cef70e844cc291f9a4e55fc9d4f9be4 | 07 September 2020, 18:52:09 UTC |
| 77792eb | Malcolm Reynolds | 07 September 2020, 18:38:15 UTC | Internal change PiperOrigin-RevId: 330402541 Change-Id: I37a1c7a3cf9f42635138c57ad0eb26dce3d580d3 | 07 September 2020, 18:38:46 UTC |
| 2827d95 | Adrian Kuegel | 24 July 2020, 07:21:12 UTC | Change saved_model_test back to use assertAllClose. We are dealing here with floating point values, so some small errors can happen. PiperOrigin-RevId: 322948216 Change-Id: I4dc8e7f728f6e3500bbf5546bbba2b418d14b13f | 24 July 2020, 07:21:44 UTC |
| 74be631 | Tom Hennigan | 16 July 2020, 11:11:35 UTC | Ensure data dependence on input to saved model. PiperOrigin-RevId: 321539026 Change-Id: If585d4dfeb8d481e966c39721ef41e4cba37799c | 16 July 2020, 11:12:04 UTC |
| 45b2b15 | Karthikeyan Singaravelan | 09 July 2020, 14:36:59 UTC | Import ABC from collections.abc | 09 July 2020, 14:36:59 UTC |
| be8e9eb | Tom Hennigan | 02 July 2020, 08:50:10 UTC | Rename whitelist to allow. PiperOrigin-RevId: 319373568 Change-Id: If0a500774fef2a67e7bef4ed8b63f976c8b038e2 | 02 July 2020, 08:50:39 UTC |
| 089e3b0 | Ruoxin Sang | 01 July 2020, 17:50:11 UTC | Update TPUStrategy symbol in Sonnet. PiperOrigin-RevId: 319255469 Change-Id: I5cf0d3596cc5e2ffffc4b93ae8184fa2d3f43dd6 | 01 July 2020, 17:50:46 UTC |
| ae34a78 | Tom Hennigan | 14 May 2020, 13:01:33 UTC | Use tfds for Cifar10 in VQ-VAE example, default to EMA and include cell outputs. PiperOrigin-RevId: 311517163 Change-Id: Ie8ebc0f95c14d6e70118f48695be65ad7773a13b | 14 May 2020, 13:02:05 UTC |
| 59f8a46 | Tom Hennigan | 09 May 2020, 11:52:43 UTC | Run tests with TensorFlow 2.2.0. PiperOrigin-RevId: 310708598 Change-Id: I7f4b141b4354931d6d241762b9419eaed1b46d46 | 09 May 2020, 11:53:07 UTC |
| 61b97be | Sonnet Contributor | 04 May 2020, 15:17:28 UTC | Change BatchApply to accept any callable that returns a tf.Tensor. PiperOrigin-RevId: 309746386 Change-Id: I6e7331e9e7facd58f8f5e87e1ba6037bd26e0c2c | 04 May 2020, 15:17:57 UTC |
| 550d8c6 | Tom Hennigan | 24 April 2020, 20:00:57 UTC | Use sphinx_autodoc_typehints for python types. PiperOrigin-RevId: 308309638 Change-Id: Ie8d2adb67ba0d3159dda3ca255b0b02c0a36d64b | 24 April 2020, 20:01:37 UTC |
| 5c5ef16 | Tom Hennigan | 20 April 2020, 13:48:57 UTC | Drop usages of `tf.enable_v2_behavior()`, no longer necessary. PiperOrigin-RevId: 307389325 Change-Id: Id5d59c4303dd6c10c0328f801051e95de1bdab23 | 20 April 2020, 13:49:25 UTC |
| 4633390 | Copybara-Service | 17 April 2020, 18:23:15 UTC | Merge pull request #168 from ialong:patch-1 PiperOrigin-RevId: 307079951 Change-Id: Ic91440e5fe76aec3cf46d39212b4dd4155a42589 | 17 April 2020, 18:23:15 UTC |
| a06702c | Tom Hennigan | 17 April 2020, 12:53:37 UTC | Update distributed training notes. PiperOrigin-RevId: 307028907 Change-Id: I5697668bf11f0037d0b580b79973f6904bad4222 | 17 April 2020, 12:54:05 UTC |
| baf0c57 | Tom Hennigan | 17 April 2020, 08:19:09 UTC | Use latest version of Sonnet in colab examples. Fixes #169. PiperOrigin-RevId: 307003064 Change-Id: I17ac33a29c2dd88ff38d2933adf80c04b2fd27c5 | 17 April 2020, 08:19:41 UTC |
| 018b05e | Alessandro Davide Ialongo | 15 April 2020, 16:46:32 UTC | add float64 test for VQ-VAE make sure `VectorQuantizerEMA` works if the dtype is tf.float64 | 15 April 2020, 16:46:32 UTC |
| 58d9a27 | Alessandro Davide Ialongo | 12 April 2020, 12:31:41 UTC | fix type of one hot indicators in vqvae.py `tf.one_hot` defaults to `tf.float32` if no dtype is provided. This prevents `VectorQuantizerEMA` from running if its dtype is specified as `tf.float64` at init time (since the EMA tries to subtract a `tf.float32` - the one hot indicator tensor - from a `tf.float64`). This is easily fixed by passing a dtype argument to `tf.one_hot`. | 12 April 2020, 12:31:41 UTC |
| 941f7b3 | Mark Daoust | 03 April 2020, 18:07:54 UTC | Remove `%tensorflow_version` and `from __future__` PiperOrigin-RevId: 304652978 Change-Id: I4e46902b49b71a98cd3ebc13fd5eb20e4c076dbf | 03 April 2020, 18:08:20 UTC |
| 3b8f9d4 | Tamara Norman | 02 April 2020, 14:13:06 UTC | Fix for change to conv_transpose PiperOrigin-RevId: 304394090 Change-Id: I7cdbfde46fc70a7c324ac8d7b4fc50ef89516639 | 02 April 2020, 14:13:31 UTC |
| f28eb90 | Tamara Norman | 02 April 2020, 12:48:58 UTC | Support shape changing inputs PiperOrigin-RevId: 304382117 Change-Id: I8b2f86adc492365c3327770aa69db01c62c1aebc | 02 April 2020, 12:49:27 UTC |
| 67a0955 | John Aslanides | 27 March 2020, 11:08:06 UTC | Update README to reflect 2.0 release. PiperOrigin-RevId: 303301305 Change-Id: Ifdef58d5cf7170da8b15bccaea7a2b8d692ec1ae | 27 March 2020, 11:08:37 UTC |
| 940b4d1 | Tom Hennigan | 27 March 2020, 10:05:43 UTC | Update Sonnet version to 2.0.0. PiperOrigin-RevId: 303292840 Change-Id: I0118990892ac9c369c94de966a092f5056d1eafd | 27 March 2020, 10:06:14 UTC |
| bb53c63 | Ken Franko | 19 March 2020, 23:32:05 UTC | Apply name change(experimental_run_v2 -> run) for all callers. PiperOrigin-RevId: 301919882 Change-Id: I14c6ed85bdf50d619d1bc572e0fbcc5f1821c70b | 19 March 2020, 23:32:32 UTC |
| de30fda | Tamara Norman | 16 March 2020, 12:28:44 UTC | Ensure batch norm works for inference when the number of channels is 1 PiperOrigin-RevId: 301137212 Change-Id: I8c9d03dce81d7b38b2d082873c9fba75662541b3 | 16 March 2020, 12:29:09 UTC |
| 7a70496 | Sonnet Copybara | 09 March 2020, 14:43:46 UTC | Merge pull request #164 from hanbyul-kim:hotfix/resolve_doc_warnings PiperOrigin-RevId: 299838560 Change-Id: I8698eed36a2b122fdfc11a83a92925da8ae5a2fb | 09 March 2020, 14:43:46 UTC |
| 733d6bb | Tamara Norman | 27 February 2020, 14:56:39 UTC | Fix typing of cifar10_convnet after previous change PiperOrigin-RevId: 297585986 Change-Id: Icdcfbdec09feafb1026385bf794d252a626a8219 | 27 February 2020, 14:57:06 UTC |
| 3debabb | Tamara Norman | 27 February 2020, 14:12:47 UTC | Internal change PiperOrigin-RevId: 297579572 Change-Id: Id57c06efd031bae6449ae02f4eb361e5ddacde38 | 27 February 2020, 14:13:15 UTC |
| 2a8e1d8 | Sonnet Contributor | 26 February 2020, 19:45:26 UTC | Explicitly create names for batch_norm moving statistics. PiperOrigin-RevId: 297411469 Change-Id: I6c9c37958930614bf47f10edf51dafb161b2ace3 | 26 February 2020, 19:46:05 UTC |
| c8fbd10 | Hanbyul Kim | 26 February 2020, 14:15:04 UTC | Fix indentation error | 26 February 2020, 14:15:04 UTC |
| 8574df2 | Hanbyul Kim | 26 February 2020, 13:35:21 UTC | Split a line into two and remove indent | 26 February 2020, 13:39:04 UTC |
| e7f7b61 | Hanbyul Kim | 26 February 2020, 13:34:00 UTC | Remove extra newlines | 26 February 2020, 13:39:04 UTC |
| 623a88d | Hanbyul Kim | 26 February 2020, 13:19:04 UTC | Fix code block to use :: | 26 February 2020, 13:39:04 UTC |
| 2095e03 | Hanbyul Kim | 25 February 2020, 14:46:58 UTC | Fix broken code block | 26 February 2020, 13:39:04 UTC |
| dab662d | Hanbyul Kim | 25 February 2020, 14:31:31 UTC | Fix unexpected indent of docstring | 26 February 2020, 13:39:04 UTC |
| d0cc578 | Sonnet Contributor | 26 February 2020, 12:32:43 UTC | In preparation for cl/296449699, disabling some pytype checks that fail after this CL is submitted. The root cause for these failures is that currently tensorflow.compat.v1 maps to Any and effectively there isn't any type checking. cl/296449699 enables this type checking for tensorflow.compat.v1 and as a result exposes some bugs / problems with the type checking code that got checked in previously. PiperOrigin-RevId: 297331371 Change-Id: I7bb4ccb53ef6aebcae69e809ce245b1f02882ea9 | 26 February 2020, 12:33:18 UTC |
| 3732f19 | Sonnet Contributor | 24 February 2020, 13:30:51 UTC | DepthwiseConv2D is now accessible via snt.DepthwiseConv2D PiperOrigin-RevId: 296870777 Change-Id: Ie1ac45cfbb8cee5b004ddffe34568044405245dc | 24 February 2020, 13:31:20 UTC |
| 1f5c0c2 | Sonnet Copybara | 21 February 2020, 09:25:53 UTC | Merge pull request #159 from chris-chris:feature/2002-mac-test PiperOrigin-RevId: 296389674 Change-Id: Ifce8e017fbbce4617657f958e83667e8d71e51bd | 21 February 2020, 09:25:53 UTC |
| d4f38d4 | Sergei Lebedev | 20 February 2020, 20:09:08 UTC | Fixed a few formatting quirks in recurrent docs PiperOrigin-RevId: 296267592 Change-Id: I9094ae88cf8ef16ee3721c8bd8f19532008a0f13 | 20 February 2020, 20:09:55 UTC |
| 4cbbe29 | Hoyean Song | 20 February 2020, 19:37:05 UTC | CPU count on macos | 20 February 2020, 19:37:05 UTC |
| 12c3b22 | Ken Franko | 19 February 2020, 18:05:28 UTC | Remove values property from DistributedValues. PiperOrigin-RevId: 295994651 Change-Id: I663f2b2b94a04e4a9ac01ca63629d7bbf412ea15 | 19 February 2020, 18:05:49 UTC |
| b6152b7 | Ken Franko | 18 February 2020, 22:56:03 UTC | Make primary property on DistributedValue private. PiperOrigin-RevId: 295829087 Change-Id: I5c0bf035780f92de197578c979301ac900dc7afc | 18 February 2020, 22:56:38 UTC |
| 4bbbcbc | Mehdi S. M. Sajjadi | 10 February 2020, 12:28:06 UTC | Make snt.Linear compatible with more than one batch dimension. The Linear layer should be compatible with inputs of dimension 2 or more. All dimensions but the last one will be treated as batch dimensions. PiperOrigin-RevId: 294193315 Change-Id: Ia673936f09832dabfaea02e612f34a59c8da187e | 10 February 2020, 12:28:28 UTC |
| f9cc3e9 | Sonnet Copybara | 30 January 2020, 14:18:49 UTC | Merge pull request #157 from deepmind:dependabot/pip/tensorflow-2.0.1 PiperOrigin-RevId: 292334684 Change-Id: Ib2fff7befb0c83110abe292dcdfbe144a549f41b | 30 January 2020, 14:18:49 UTC |
| 9eaac57 | dependabot[bot] | 28 January 2020, 21:46:22 UTC | Bump tensorflow from 2.0.0 to 2.0.1 Bumps [tensorflow](https://github.com/tensorflow/tensorflow) from 2.0.0 to 2.0.1. - [Release notes](https://github.com/tensorflow/tensorflow/releases) - [Changelog](https://github.com/tensorflow/tensorflow/blob/master/RELEASE.md) - [Commits](https://github.com/tensorflow/tensorflow/compare/v2.0.0...v2.0.1) Signed-off-by: dependabot[bot] <support@github.com> | 28 January 2020, 21:46:22 UTC |
| f18c55e | Tom Hennigan | 27 January 2020, 14:23:29 UTC | Fuzz tests for more optimizers. PiperOrigin-RevId: 291709836 Change-Id: Ia1282ba02649863e1ea39c481a84e31f68ace7ea | 27 January 2020, 14:23:57 UTC |
| 4fe2373 | Sonnet Copybara | 27 January 2020, 09:46:12 UTC | Merge pull request #155 from fostiropoulos:v2 PiperOrigin-RevId: 291678292 Change-Id: I390276d492db8e7b1aaaf994b9e8592391d2e7ad | 27 January 2020, 09:46:12 UTC |
| 82049c2 | fostiropoulos | 25 January 2020, 19:48:24 UTC | Fixed vqvae_example.ipynb to be compatible with tf v2 | 25 January 2020, 19:48:24 UTC |
| c616bbd | Tom Hennigan | 22 January 2020, 20:12:17 UTC | Expose `snt.{merge,split}_leading_dims`. PiperOrigin-RevId: 291003968 Change-Id: I7678d4aa15bb08650203afe0e21da9141c96a574 | 22 January 2020, 20:12:43 UTC |
| af606a4 | Tom Hennigan | 22 January 2020, 09:52:41 UTC | Shard conformance tests to allow faster testing. PiperOrigin-RevId: 290911624 Change-Id: Iacf31a4798c428942708acc43ecf0d9f071ce88c | 22 January 2020, 09:53:17 UTC |
| f44c133 | Tom Hennigan | 19 January 2020, 13:16:26 UTC | Create `snt.distribute.create_variables_eagerly(f)`. `create_variables_eagerly` is a higher order function that attempts to lift all variable creation within a given function into eager mode. It does this by patching Sonnet initializers to run under a `tf.init_scope`, and then patching variable creation to also run under a `tf.init_scope`. Additionally we support statically determining some values not created using Sonnet (e.g. `tf.zeros`). This function is useful when doing large scale distributed training with TensorFlow on clusters of machines. We can avoid much of the checking that happens for variables created in tf.functions which result in a lot of RPCs between the coordinator and workers, slowing down the first training step. PiperOrigin-RevId: 290499884 Change-Id: I1fce83011df198ff23c044689b6e5d0e60094be2 | 19 January 2020, 13:16:53 UTC |
| 25f040a | Sonnet Contributor | 17 January 2020, 10:33:27 UTC | Fix bug in _rnn_step, by swapping the prev_state with state. PiperOrigin-RevId: 290235436 Change-Id: Iea5fb1d0b6a14e6f050212293861b9fc107b7e9c | 17 January 2020, 10:34:00 UTC |
| d8d1652 | Chris Jones | 15 January 2020, 14:40:14 UTC | Don't run stateful mixed precision example in doctest. PiperOrigin-RevId: 289848542 Change-Id: I66743dd05dccdac8a8b063b73995601a6c430652 | 15 January 2020, 14:40:38 UTC |
| 0c623c0 | Tom Hennigan | 08 January 2020, 13:04:39 UTC | Expose block group and bottle neck blocks from resnet. PiperOrigin-RevId: 288675258 Change-Id: Id3eb4de377f09a497fb5cfd680d7f898e4413e3a | 08 January 2020, 13:05:11 UTC |
| 431c6e4 | Tamara Norman | 07 January 2020, 15:20:02 UTC | Add check on inputs and outputs for TPU strategy test PiperOrigin-RevId: 288491883 Change-Id: I6f1b8db69ca360d152bf6eeb21bf1c06aeedbf92 | 07 January 2020, 15:20:57 UTC |
| 6e5de35 | Ken Franko | 06 January 2020, 16:54:31 UTC | Add cross-replica BatchNorm test on TPU strategy. PiperOrigin-RevId: 288309269 Change-Id: Ibe831d68d273fd9d279e663625d1fed7e84bba46 | 06 January 2020, 16:55:03 UTC |
| 1d16fc1 | Sonnet Contributor | 02 January 2020, 18:20:12 UTC | Remove eager variable creation hack. Underlying performance issue has been fixed. Workaround no longer needed. PiperOrigin-RevId: 287854778 Change-Id: Idc9609bf73baa432700bc2fe69dafefdafd970c6 | 02 January 2020, 18:20:45 UTC |
| 361f033 | Tom Hennigan | 02 January 2020, 09:05:41 UTC | Enable ResNet50 V2, has reproduced canonical experiment. PiperOrigin-RevId: 287803490 Change-Id: Id59f82145be188d1f587602d604c5aa12e9f5e2d | 02 January 2020, 09:06:13 UTC |
| d7f1998 | Sergei Lebedev | 18 December 2019, 10:34:29 UTC | Removed wrapt.ObjectProxy hack from TrainableState tree now supports wrapped objects in the same way as tf.nest PiperOrigin-RevId: 286154954 Change-Id: I35dfeedf96be5d85bb1b74e2d79961e91172175b | 18 December 2019, 10:34:57 UTC |
