training: v0.9.0

0.9.0 (2026-01-28)

⚠ BREAKING CHANGES

• models,training: multi dataset integration

Features

• Drop python 3.10 (#795) (ebea6d0)
• models,training: Multi dataset integration (f537d7f)
• training: Autoencoder 🗜️ (#252) (50e497b)

Bug Fixes

• deps: Workaround for failing uv dependency resolution (#828) (bc64bac)
• Set min version for azureml-mlflow (#842) (0f721a5)

models: v0.12.0

0.12.0 (2026-01-28)

⚠ BREAKING CHANGES

• models,training: multi dataset integration

Features

• Drop python 3.10 (#795) (ebea6d0)
• models, training: Expose row_normalize kwarg to TruncatedConnectionSchema (#805) (20b2da6)
• models,training: Multi dataset integration (f537d7f)
• training: Autoencoder 🗜️ (#252) (50e497b)

graphs: v0.8.4

0.8.4 (2026-01-28)

Features

• Drop python 3.10 (#795) (ebea6d0)
• graphs: Introduce a new multiscale edgebuilding strategy optimizing 1-hop edges (#694) (2385b9d)

Bug Fixes

• graphs: Typo in LimitedAreaNPZNodes (#823) (b267f42)
• System level tests (#812) (a402373)

Documentation

• Update tri_nodes.csv - refinement lvl 9 (#807) (2b626ca)

training: v0.8.4

0.8.4 (2026-01-14)

Features

• Introduce spectral losses module (#678) (9745b46)
• models: Mapper refactor (#574) (83820c8)

Bug Fixes

• graphs,aicon: Integration test (#794) (1d8770c)
• Remove limit on numpy version (#688) (bc6c6cb)
• training: Dataloader worker ranges (#764) (3ce36ae)

models: v0.11.3

0.11.3 (2026-01-14)

Features

• Introduce spectral losses module (#678) (9745b46)
• models: Mapper refactor (#574) (83820c8)

Bug Fixes

• models: Fix TruncatedConnection schema and initialization (#787) (c24c195)
• training: Dataloader worker ranges (#764) (3ce36ae)

graphs: v0.8.3

0.8.3 (2026-01-14)

Bug Fixes

• Remove limit on numpy version (#688) (bc6c6cb)

training: v0.8.3

0.8.3 (2026-01-05)

Bug Fixes

• ensemble training: Training is broken when ensemble is sharded across gpus (#774) (ca08316)
• Reverting change to include dependency of mlflow-export-import to unblock pypi releases (#775) (f7d2acc)
• tests: Expand losses test suite (#754) (e6b7dbb)
• training: Convert warm_start_path to pathlib.Path object (#751) (59f033c)

training: v0.8.2

0.8.2 (2025-12-18)

Features

• Multi-scale loss implementations (including multi-scale kcrps) (#388) (17cb84d)

Bug Fixes

• docs,schemas: Update hardware reference in docstrings, documentation and schemas (#737) (4872c7c)
• Fix check for checkpoint path in the dry run case. (#755) (0f82efa)
• Mlflow-export-import new inputs format (#748) (7bca8e8)
• Progress bar with lightning==2.6.0 (#739) (e7453ae)
• training: Add missing rad2deg in plot sample (#744) (a5fedb3)
• Update the CLI to generate a MLFlow run ID. (#758) (88bf1e5)
• Wrong model for ens in benchmarking tests (#760) (a0d9984)

graphs: v0.8.2

0.8.2 (2026-01-05)

Features

• graphs: Add utility function to transform between sincos and latlon_rad (#761) (4e962d5)

Bug Fixes

• Pin Netcdf4 package to 1.7.3 as max version (#776) (0050ad9)

models: v0.11.2

0.11.2 (2025-12-18)

Features

• models: Triton gt more shapes (#752) (0608acf)

Bug Fixes

• Correct for dimensions of skipped connection in conditioning (4551e51)
• Fix-pytest-triton-error (#740) (00b38c9)
• residual: Fix conditioning on skipped connection (#742) (4551e51)