training: v0.8.1

0.8.1 (2025-12-08)

Features

• Edge pre mlp for gt conv (#695) (c72a789)
• Noise injector refactor (#724) (e1000af)

Bug Fixes

• training: Handle CombinedLoss in print_variable_scaling (#723) (b650c59)

models: v0.11.1

0.11.1 (2025-12-08)

Features

• Edge pre mlp for gt conv (#695) (c72a789)
• Noise injector refactor (#724) (e1000af)

graphs: v0.8.1

0.8.1 (2025-12-08)

Bug Fixes

• graphs,multi-scale: Set default scale_resolutions from 1 (#733) (e621fc9)
• graphs: Allow scale_resolutions to be 0 in MultiScaleEdges (#731) (11af322)

training: v0.8.0

0.8.0 (2025-12-05)

⚠ BREAKING CHANGES

• training: Refactor configuration by introducing system schema with hardware, paths, and files subschemas (#598)
• cond layer norm (#658)

Features

• Activate minmium plotting for integration tests (#669) (84e5882)
• Compile transformer gnn (#181) (24d162c)
• models: Add configurable residual connections in enc-proc-dec (#670) (aeaf00b)
• models: Triton GraphTransformer (#631) (b40b6c6)
• Time_interpolator_callbacks (#677) (c2b8179)
• training: Performance docs (#696) (9574ff1)
• training: Refactor optimizer creation to support custom and torch optimizers (#588) (cd777fb)

Bug Fixes

• Add package config path to Hydra search path in plugin (#656) (ca6f732)
• Cond layer norm (#658) (7315e3a)
• logger: Bugs in AzureMLFlowLogger from #646 (#685) (14c0235)
• models: Processor chunking (#629) (06e5533)
• Pass weights_only for pytorch lightning >= 2.6.0 (#713) (7446942)
• Ptl 2.6.0 explicitly pass weights_only=False (#710) (e18824c)
• RolloutEval sharding (#714) (0fbc071)
• Slurm system config (#702) (cce8763)
• Target docs (#704) (200101e)
• training,tasks: Abstract RolloutForecasting task (#682) (f14fc32)
• training: CombinedLoss schema validation (#719) (dba4268)
• training: Refactor configuration by introducing system schema with hardware, paths, and files subschemas (#598) (da02fe7)
• training: Remove unused code (#706) (f49813a)

models: v0.11.0

0.11.0 (2025-12-05)

⚠ BREAKING CHANGES

• training: Refactor configuration by introducing system schema with hardware, paths, and files subschemas (#598)
• cond layer norm (#658)

Features

• Compile transformer gnn (#181) (24d162c)
• models: Add configurable residual connections in enc-proc-dec (#670) (aeaf00b)
• models: Multibackend all_to_all wrapper (#95) (6819be1)
• models: Triton GraphTransformer (#631) (b40b6c6)

Bug Fixes

• Compile pickle error (#708) (f4fc4ab)
• Cond layer norm (#658) (7315e3a)
• models: Processor chunking (#629) (06e5533)
• Predict_step shard shapes (#692) (be9ff8b)
• Remove import of anemoi training in compile (#705) (f7d5ae4)
• Small pytorch boxcox inefficiency (#683) (66b40e0)
• training: Refactor configuration by introducing system schema with hardware, paths, and files subschemas (#598) (da02fe7)

graphs: v0.8.0

0.8.0 (2025-12-05)

⚠ BREAKING CHANGES

• edges: Edge feature revision #643 (#727)

Features

• edges: Edge feature revision (#643) (720f4d8)
• edges: Edge feature revision #643 (#727) (d1372cf)
• graphs: Support for multi-scale connections with HEALPix hidden grid (#691) (1450787)

Bug Fixes

• Revert "feat(edges): Edge feature revision" (#726) (db1f940)
• Sparse export (#686) (969b787)
• Target docs (#704) (200101e)

training: v0.7.0

0.7.0 (2025-11-17)

⚠ BREAKING CHANGES

• training: remove support for EDA (#651)

Features

• Callbacks for GraphEnsForecaster. (#449) (39c2bfc)
• Mlflow azure (#646) (27bd3dd)
• training: Remove support for EDA (#651) (921e108)

Bug Fixes

• Anemoi-datasets import (#626) (65c8901)
• Bug for mlflow offline logging (#675) (fdce0f6)
• Bug in sample plots. (#632) (9e024f3)
• Incorrect test for variable mask scaler (#649) (d0f775e)
• Integration tests and drop missed reference for profiler (#630) (f352e17)
• training: Provide more informative error when user specifies inexistent node attribute (#663) (dde3cb6)
• Update readmes (#655) (a58aa64)

models: v0.10.0

0.10.0 (2025-11-17)

⚠ BREAKING CHANGES

• training: remove support for EDA (#651)

Features

• training: Remove support for EDA (#651) (921e108)

Bug Fixes

• Basemodel.predict_step (#672) (0c830e9)
• models: Assert no dropout (#638) (c1bbcec)
• Shard shape type hints (#625) (fb201fd)
• Update readmes (#655) (a58aa64)

graphs: v0.7.2

0.7.2 (2025-11-17)

Features

• graphs: Add LimitedAreaMask for stretched hidden nodes (#671) (f155f3c)
• graphs: New edge attributes and faster graph cleaning (#617) (8659de9)
• New interactive graph visualization (#609) (5b5ede4)

Bug Fixes

• graphs,normalisation: Add assert when dividing by 0 (#676) (01b7034)
• graphs,schemas: Missing type for mask_attr_name in schema (#664) (f021017)
• graphs,tests: New test and fix anemoi-graphs tests with gpu (#637) (ca1b542)
• graphs: Remove wrong argument from og.Figure (#616) (abd37eb)
• graphs: Unit-range normalisation (#665) (6de4778)
• Update readmes (#655) (a58aa64)

training: v0.6.7

0.6.7 (2025-10-20)

Features

• models: Point-mlp proccesor (#367) (ee2a067)

Bug Fixes

• CPU integration test - update AICON test to use anemoi-utils data download fixtures (#608) (4c53c95)
• Refactored paths to import from package directly (#601) (f28bfd4)
• training: Add default empty forcing list to DataSchema (#602) (18a609b)
• training: Nan mask weights gather for plots (#537) (a293b99)

Documentation

• training: More explicit comment about order of preprocessors in config file (#613) (bbd152d)