Skip to content

Releases: LukasHedegaard/ride

Library compatibility updates

17 Apr 16:04
cce91b7
Compare
Choose a tag to compare

Fixed

  • Various minor details to fix compatibility with newer versions of used libraries

Fix requirements

03 Jun 09:36
8ea3dc1
Compare
Choose a tag to compare

Added

  • Version for protobuf during build.
  • Conditional install of redis on win platforms

Fix missing device transfer in benchmark

18 Mar 11:37
b3aa999
Compare
Choose a tag to compare
Merge pull request #16 from LukasHedegaard/develop

Fix missing device transfer in benchmark

Improved profiling, new MultiStepLR optimizer, and minor fixes

18 Mar 10:38
20836c6
Compare
Choose a tag to compare

Added

  • Defensive fallback for FLOPs measurement.
  • Add MultiStepLR optimizers.

Changed

  • Profiling to use pytorch_benchmark package.

Fixed

  • WandB logger log_dir extraction.

Improved profiling

07 Feb 14:57
69b3397
Compare
Choose a tag to compare

Changed

  • Profile only warms up on first inference.

Memory profiling

27 Jan 15:07
745c008
Compare
Choose a tag to compare

Added

  • Memory profiling.

Fixed

  • Tune DeprecationWarning.

Fix missing gradient on train_step

16 Nov 12:39
31f707c
Compare
Choose a tag to compare

What's Changed

Added

  • Add pred and target dict support in Lifecycle.

Fixed

  • Avoid detaching loss in step.

Full Changelog: 0.5.0...0.5.1

Add MetricSelector, preprocess_batch, fix minor issues and update PL and Tune versions

12 Nov 08:21
08b24b3
Compare
Choose a tag to compare

Added

  • Add preprocess_batch method to Lifecycle.
  • Add option for string type in utils.name.
  • Add MetricSelector.

Fixed

  • Weight freezing during model loading.
  • Fix discriminative_lr param selection for NoneType parameters.
  • Fix wandb project naming during hparamsearch.
  • Optimizer Schedulers take accumulate_grad_batches into account.

Changed

  • Key debug statements while loading models to include both missing and unexpected keys.
  • Bumped PL to version 1.4. Holding back on 1.5 due to Tune integration issues.
  • Bumped Tune to version 1.8.

What's Changed

Full Changelog: 0.4.6...0.5.0

Update profile and allow warm_up overloading in mixins

21 Sep 09:18
e3851fd
Compare
Choose a tag to compare

Fixed

  • Update profile to use model.call. This enable non-forward executions during profiling.
  • Add DefaultMethods Mixin with warm_up to make warm_up overloadable by Mixins.

Minor fixes in warm_up and 3rd-party versioning

08 Sep 12:25
fe979b8
Compare
Choose a tag to compare

Fixed

  • Fix warm_up function signature
  • Requirement versions