|
5 | 5 | Change Log |
6 | 6 | ========== |
7 | 7 |
|
| 8 | +Release 2.6 - 1/19/2022 |
| 9 | +----------------------- |
| 10 | + |
| 11 | +**NOTE**: NNI v2.6 is the last version that supports Python 3.6. From next release NNI will require Python 3.7+. |
| 12 | + |
| 13 | +Hyper-Parameter Optimization |
| 14 | +^^^^^^^^^^^^^^^^^^^^^^^^^^^^ |
| 15 | + |
| 16 | +Experiment |
| 17 | +"""""""""" |
| 18 | + |
| 19 | +* The legacy experiment config format is now deprecated. `(doc of new config) <https://nni.readthedocs.io/en/v2.6/reference/experiment_config.html>`__ |
| 20 | + |
| 21 | + * If you are still using legacy format, nnictl will show equivalent new config on start. Please save it to replace the old one. |
| 22 | + |
| 23 | +* nnictl now uses ``nni.experiment.Experiment`` `APIs <https://nni.readthedocs.io/en/stable/Tutorial/HowToLaunchFromPython.html>`__ as backend. The output message of create, resume, and view commands have changed. |
| 24 | +* Added Kubeflow and Frameworkcontroller support to hybrid mode. `(doc) <https://nni.readthedocs.io/en/v2.6/TrainingService/HybridMode.html>`__ |
| 25 | +* The hidden tuner manifest file has been updated. This should be transparent to users, but if you encounter issues like failed to find tuner, please try to remove ``~/.config/nni``. |
| 26 | + |
| 27 | +Algorithms |
| 28 | +"""""""""" |
| 29 | + |
| 30 | +* Random tuner now supports classArgs ``seed``. `(doc) <https://nni.readthedocs.io/en/v2.6/Tuner/RandomTuner.html>`__ |
| 31 | +* TPE tuner is refactored: `(doc) <https://nni.readthedocs.io/en/v2.6/Tuner/TpeTuner.html>`__ |
| 32 | + |
| 33 | + * Support classArgs ``seed``. |
| 34 | + * Support classArgs ``tpe_args`` for expert users to customize algorithm behavior. |
| 35 | + * Parallel optimization has been turned on by default. To turn it off set ``tpe_args.constant_liar_type`` to ``null`` (or ``None`` in Python). |
| 36 | + * ``parallel_optimize`` and ``constant_liar_type`` has been removed. If you are using them please update your config to use ``tpe_args.constant_liar_type`` instead. |
| 37 | + |
| 38 | +* Grid search tuner now supports all search space types, including uniform, normal, and nested choice. `(doc) <https://nni.readthedocs.io/en/v2.6/Tuner/GridsearchTuner.html>`__ |
| 39 | + |
| 40 | +Neural Architecture Search |
| 41 | +^^^^^^^^^^^^^^^^^^^^^^^^^^ |
| 42 | + |
| 43 | +* Enhancement to serialization utilities `(doc) <https://nni.readthedocs.io/en/v2.6/NAS/Serialization.html>`__ and changes to recommended practice of customizing evaluators. `(doc) <https://nni.readthedocs.io/en/v2.6/NAS/QuickStart.html#pick-or-customize-a-model-evaluator>`__ |
| 44 | +* Support latency constraint on edge device for ProxylessNAS based on nn-Meter. `(doc) <https://nni.readthedocs.io/en/v2.6/NAS/Proxylessnas.html>`__ |
| 45 | +* Trial parameters are showed more friendly in Retiarii experiments. |
| 46 | +* Refactor NAS examples of ProxylessNAS and SPOS. |
| 47 | + |
| 48 | +Model Compression |
| 49 | +^^^^^^^^^^^^^^^^^ |
| 50 | + |
| 51 | +* New Pruner Supported in Pruning V2 |
| 52 | + |
| 53 | + * Auto-Compress Pruner `(doc) <https://nni.readthedocs.io/en/v2.6/Compression/v2_pruning_algo.html#auto-compress-pruner>`__ |
| 54 | + * AMC Pruner `(doc) <https://nni.readthedocs.io/en/v2.6/Compression/v2_pruning_algo.html#amc-pruner>`__ |
| 55 | + * Movement Pruning Pruner `(doc) <https://nni.readthedocs.io/en/v2.6/Compression/v2_pruning_algo.html#movement-pruner>`__ |
| 56 | + |
| 57 | +* Support ``nni.trace`` wrapped ``Optimizer`` in Pruning V2. In the case of not affecting the user experience as much as possible, trace the input parameters of the optimizer. `(doc) <https://nni.readthedocs.io/en/v2.6/Compression/v2_pruning_algo.html>`__ |
| 58 | +* Optimize Taylor Pruner, APoZ Activation Pruner, Mean Activation Pruner in V2 memory usage. |
| 59 | +* Add more examples for Pruning V2. |
| 60 | +* Add document for pruning config list. `(doc) <https://nni.readthedocs.io/en/v2.6/Compression/v2_pruning_config_list.html>`__ |
| 61 | +* Parameter ``masks_file`` of ``ModelSpeedup`` now accepts `pathlib.Path` object. (Thanks to @dosemeion) `(doc) <https://nni.readthedocs.io/en/v2.6/Compression/ModelSpeedup.html#user-configuration-for-modelspeedup>`__ |
| 62 | +* Bug Fix |
| 63 | + |
| 64 | + * Fix Slim Pruner in V2 not sparsify the BN weight. |
| 65 | + * Fix Simulator Annealing Task Generator generates config ignoring 0 sparsity. |
| 66 | + |
| 67 | +Documentation |
| 68 | +^^^^^^^^^^^^^ |
| 69 | + |
| 70 | +* Supported GitHub feature "Cite this repository". |
| 71 | +* Updated index page of readthedocs. |
| 72 | +* Updated Chinese documentation. |
| 73 | + |
| 74 | + * From now on NNI only maintains translation for most import docs and ensures they are up to date. |
| 75 | + |
| 76 | +* Reorganized HPO tuners' doc. |
| 77 | + |
| 78 | +Bugfixes |
| 79 | +^^^^^^^^ |
| 80 | + |
| 81 | +* Fixed a bug where numpy array is used as a truth value. (Thanks to @khituras) |
| 82 | +* Fixed a bug in updating search space. |
| 83 | +* Fixed a bug that HPO search space file does not support scientific notation and tab indent. |
| 84 | + |
| 85 | + * For now NNI does not support mixing scientific notation and YAML features. We are waiting for PyYAML to update. |
| 86 | + |
| 87 | +* Fixed a bug that causes DARTS 2nd order to crash. |
| 88 | +* Fixed a bug that causes deep copy of mutation primitives (e.g., LayerChoice) to crash. |
| 89 | +* Removed blank at bottom in Web UI overview page. |
| 90 | + |
8 | 91 | Release 2.5 - 11/2/2021 |
9 | 92 | ----------------------- |
10 | 93 |
|
|
0 commit comments