v0.1.8

What's Changed

• [FEATURE] Add CrossMAE as a method by @d3tk in #423
• [Feature] Major Updates to SPT Web by @MarcelMatsal in #428
• [Feature] Add PMSN as a method by @kitewatermelon in #429

New Contributors

• @d3tk made their first contribution in #423

Full Changelog: v0.1.7...v0.1.8

v0.1.7

What's Changed

• [Doc] Update documentation for easier onboarding by @flogrammer in #399
• [Feature] Add SALT (Static-teacher Asymmetric Latent Training) pipeline by @master in #396
• Fix named_parameters() warning firing on every call by @mohsinm-dev in #389
• Allow MAE and IJEPA to accept pre-instantiated nn.Module as encoder. Support RoPE-based models (DINOv3) in MaskedEncoder by @haodongzhang0118 in #393
• [Feature] Implement VCRegLoss and integrate with VICRegLoss by @quentinll in #366
• Add Manager resume_weights_only and forward to Trainer.fit by @huguesva in #390
• [Feature] give SubSet access to dataset attr/methods by @lucas-maes in #402
• fix(docs): correct IJEPA import path in docstring example by @kitewatermelon in #407
• [Docs] Add citation section to README by @huguesva in #405
• [FIX] Inconsistent/incorrect import paths in examples(MAE, LeJEPA) by @kitewatermelon in #410
• [Doc] Optuna HP optimization by @flogrammer in #412
• EnvironmentDumpCallback: drop _dump_thread when pickled by @AyushExel in #418
• [BUG] fix two spawn-mode pickle bugs in Subset and DatasetMixin by @AyushExel in #419
• fix(module): accept functools.partial as forward in spawn workers (#417) by @AyushExel in #420
• [Docs] Agent optimization by @MarcelMatsal in #422

New Contributors

• @flogrammer made their first contribution in #399
• @master made their first contribution in #396
• @haodongzhang0118 made their first contribution in #393
• @kitewatermelon made their first contribution in #407
• @AyushExel made their first contribution in #418

Full Changelog: v0.1.6...v0.1.7

v0.1.6

What's Changed

• [Test] Add test for video transforms by @lucas-maes in #372
• [BUGFIX] MultiViewTransform by @sami-bg in #375
• [FEATURE] exclude_bias_norm global flag by @sami-bg in #376
• [BUGFIX] Prior MultiViewTransforms bugfix breaks DINO and DINO2 forwards in forward.py by @sami-bg in #377
• [Feature] I-JEPA + MAE benchmarks on ImageNet by @sami-bg in #387
• [Feature] LeJEPA INet10 Single GPU benchmark by @sami-bg in #394

Full Changelog: v0.1.5...v0.1.6

v0.1.5

What's Changed

• [Feature] vit-huge by @sami-bg in #352
• Fix OnlineProbe docstring to document LARS optimizer default by @mohsinm-dev in #355
• [BUGFIX] Update pyproject.toml by @sami-bg in #354
• Fix AMP compatibility with multiple optimizers by @mohsinm-dev in #356
• [BugFix] Fix failing unit tests on main branch by @huguesva in #357
• [Feature] updated toml versions by @RandallBalestriero in #358
• [DOC] Add github action for doc update by @lucas-maes in #359
• [DOC] Update readme by @lucas-maes in #360
• [Feature] Add automatic loss logging to all forward functions by @huguesva in #361
• [BugFix] Add missing module parameter to OnlineProbe callbacks by @huguesva in #363
• [Feature] module-wide log by @RandallBalestriero in #369
• Delete stable_pretraining/_version.py by @lucas-maes in #370

New Contributors

• @mohsinm-dev made their first contribution in #355

Full Changelog: v0.1.4...v0.1.5

v0.1.3

What's Changed

• [Feature] Add flexible CLI with smart multirun detection for SLURM jobs by @huguesva in #306
• [BugFix] gallery CI fix by @huguesva in #307
• [BugFix] ema coefficient as buffer by @huguesva in #308
• [Feature] more flexible multi-view transform by @huguesva in #311
• [Feature] dict structure for MultiViewTransform by @huguesva in #312
• [Feature] DINOv1 by @huguesva in #309
• [Feature] weight decay scheduler callback by @RandallBalestriero in #314
• [BUGFIX] Manager detecting user's checkpointing intent by @sami-bg in #301
• [Benchmark] data loader : multiview vs roundrobin by @huguesva in #315
• [BugFix] Teacher Student and ReturnFeatures by @RandallBalestriero in #317
• [Feature] iBot Loss, Sinkhorn-Knopp & losses folder organization by @huguesva in #320
• [Feature] DINOv2 by @huguesva in #321
• [MRG] SwAV Implementation by @luizfysouza in #288
• [BugFix] Fix queue callbacks to handle batch size larger than queue length by @huguesva in #324
• [Feature] cleanup callback + probes by @RandallBalestriero in #326
• [Benchmark] Additional Resnet50 Benchmarking on ImageNet-100 by @wenquanlu in #327
• [BugFix] Fix callback parameter name: use 'module' instead of 'pl_module' by @huguesva in #329
• [BugFix] Fix in channels full resolution by @huguesva in #330
• [BugFix] Make instantiation warnings and config printing rank-zero-only by @huguesva in #331
• [Feature] detach utils for callbacks by @RandallBalestriero in #334
• [Feature] EnvCallback by @RandallBalestriero in #335
• [Feature] Add matmul_precision config parameter for TF32 control by @huguesva in #336
• [Feature] Add eval resolver for OmegaConf by @huguesva in #338
• [BugFix] Fix module config interpolation resolution by @huguesva in #339
• [BugFix] no optimizer module should return state by @lucas-maes in #340
• [Feature] Threaded reader + GDrive utils by @RandallBalestriero in #343
• [Doc] Google DocStyle by @lucas-maes in #344
• Test module by @quentinll in #345
• [BugFix] pipy release by @lucas-maes in #348
• [BugFix] Remove version.py by @lucas-maes in #350

New Contributors

• @wenquanlu made their first contribution in #327
• @quentinll made their first contribution in #345

Full Changelog: v0.1.2...v0.1.3

v0.1.0

v0.1.0