v0.10.8-alpha

• Bug fix: Error handling in computation of best metric value (min / max) for cases when the metric / sub-trial returned a non-numeric type (e.g. str / Exception)
• Added flexible functionality to delete model cache after completing a trial. Functionality is non-default and triggered by cerebros.cerebros_simple_random_search.CerebrosSimpleRandomSearch .get_best_model(), to prevent cumulative disk space exhaustion or in a container, memory / RAM exhaustion.

v0.10.7-alpha

Updated the mask_zero arg based on the new tokienizer.

v0.10.6-alpha

1. Refactor the text classification example to toeknize all samples first then run train the model, rather than the model tokenzing within the training loop.
2. Add support for gradient accumulation steps.
3. 0.10.5 skipped because of 2 commits on the same day.

v0.10.4-alpha

Added proper garbage collection after each sub-trial model is trained, so [GPU] RAM pressure is not cumulative across trials.

v0.10.3-alpha

Added model serialization which allows custom objects to serialize to the NLP example.

v0.10.2-alpha

Updates to documentation based on updated test scripts for Ames benchmarks.

v0.10.1-alpha

• Improvements to readme.md to highlight O(n) / subquadratic complexity timing on NLP tasks.
• Improvements to the test scripts regression-example-ames-no-preproc.py and regression-example-ames-no-preproc-val-set.py to correct data leakage in feature selection.
• Same corrections made to realnn-regression-example-ames-no-preproc.py (moot changes, script no longer used, but for the sake of correctness, so updated)

Thank you @Shohail-Ismail for raising the issue on data leakage and pushing the update to the documentation.

v0.10.0-alpha: Merge pull request #172 from david-thrower/171-upgrade-tf-2190

Summary of changes:

• Replaced the text embedding base model with an interleaved Rotaty Positional Embedding iRoPE in the Phishing detection NLP proof of concept
• Proof of concept that the entire model can scale at O(n) timing as sequence length increases infinitely.

Benchmarks:

seq_len     val_bianry_accuracy       min/model        total_min         timing_relative_to_1024                      Commit_SHA
3072          0.955                    65.942           329.715                    2.817                     4bc217b36d1baf8b9a81fe53000e9e13e6e87801
1536          0.96                     37.27             186.36                    1.591                     286ba81a1e51493d748ded727bd602a4398248a8
1024          0.952                    23.42             117.08                    1.0                       9893bfc55d4f7b753eff79e0c5c70e4992c61085

• Upgraded tf to 2.19.0
• Upgraded Jax
• Upgraded model architecture in both phishing CICD example and CIFAR10 example to accommodate tf 2.19.0.
• Removed obsolete BERT embedding CICD test.

v0.9.9-alpha

Merge pull request #150 from david-thrower/148-tensorflow-upgrades

v0.9.8-alpha

Merge pull request #127 from david-thrower/125-add-use-cases-gpt2-enc…