2.2.2

What's Changed

• Chronos-2: Add after_batch callback by @abdulfatir in #436
• Speed up predict_df by @shchur in #437
• Bump version from 2.2.1 to 2.2.2 by @abdulfatir in #439

Full Changelog: v2.2.1...v2.2.2

2.2.1

What's Changed

• Chronos-2: Update quick start notebook with LoRA example by @abdulfatir in #415
• Add unittests for df_utils by @abdulfatir in #414
• Chronos-2: Only pin_memory when device type is cuda by @abdulfatir in #431
• Chronos-2: Add option to disable DataParallel by @abdulfatir in #434
• Update version to 2.2.1 by @abdulfatir in #435

Full Changelog: v2.2.0...v2.2.1

2.2.0

🚀 What's New?

• Support for low-rank adaption (LoRA) fine-tuning
• Optimized dataframe operations, making predict_df significantly faster than before

🐛 Bug Fixes and Enhancements

• Removed assert on batch size in multi-GPU environments
• Added option to skip dataframe validation in predict_df
• Chronos-2: Renamed predict_batches_jointly to cross_learning for clarity
• Chronos-2: Added option to specify callbacks in fit
• Chronos-2: Handle prediction_length < 3 for case where future_df is provided

What's Changed

• Chronos-2: Convert assert about batch size to warning by @abdulfatir in #392
• Fix link in Chronos-Bolt tutorial notebook entry by @shchur in #396
• Optimize convert_df_input_to_list_of_dicts_input and validate_df_inputs by @shchur in #395
• Chronos-2: Add LoRA fine-tuning support by @abdulfatir in #393
• Chronos-2: Add option to skip dataframe validation in predict_df by @abdulfatir in #400
• Fix potential crash due to unbound local in df_utils by @abdulfatir in #404
• Chronos-2: Add option to specify callbacks in fit by @abdulfatir in #405
• Handle prediction_length below 3 by @shchur in #407
• Chronos-2: set default dataloader_num_workers=0 in Trainer by @abdulfatir in #409
• Chronos-2: Add option to remove PrinterCallback by @abdulfatir in #410
• Bump version to 2.2.0rc3 by @abdulfatir in #411
• Chronos-2: Do not catch exceptions in predict_fev when fine-tuning by @abdulfatir in #417
• Chronos-2: Rename predict_batches_jointly to cross_learning by @abdulfatir in #418
• Chronos-2: Update context_length if fine-tuned with longer than default by @abdulfatir in #419
• Chronos-2: Ensure updated chronos_config is saved after fine-tuning by @abdulfatir in #420
• Chronos-2: Handle missing 'peft' and lora_config specified by @abdulfatir in #421
• Bump version from 2.2.0rc3 to 2.2.0rc4 by @abdulfatir in #422
• Bump version from 2.2.0rc4 to 2.2.0 by @abdulfatir in #426
• Add validate_inputs to predict_df in BaseChronosPipeline by @abdulfatir in #427

Full Changelog: v2.1.0...v2.2.0

2.2.0rc4

What's Changed

• Chronos-2: Do not catch exceptions in predict_fev when fine-tuning by @abdulfatir in #417
• Chronos-2: Rename predict_batches_jointly to cross_learning by @abdulfatir in #418
• Chronos-2: Update context_length if fine-tuned with longer than default by @abdulfatir in #419
• Chronos-2: Ensure updated chronos_config is saved after fine-tuning by @abdulfatir in #420
• Chronos-2: Handle missing 'peft' and lora_config specified by @abdulfatir in #421
• Bump version from 2.2.0rc3 to 2.2.0rc4 by @abdulfatir in #422

Full Changelog: v2.2.0rc3...v2.2.0rc4

2.2.0rc3

What's Changed

• Handle prediction_length below 3 by @shchur in #407
• Chronos-2: set default dataloader_num_workers=0 in Trainer by @abdulfatir in #409
• Chronos-2: Add option to remove PrinterCallback by @abdulfatir in #410
• Bump version to 2.2.0rc3 by @abdulfatir in #411

Full Changelog: v2.2.0rc2...v2.2.0rc3

2.2.0rc2

What's Changed

• Fix potential crash due to unbound local in df_utils by @abdulfatir in #404
• Chronos-2: Add option to specify callbacks in fit by @abdulfatir in #405

Full Changelog: v2.2.0rc1...v2.2.0rc2

v2.2.0rc1

What's Changed

• Chronos-2: Convert assert about batch size to warning by @abdulfatir in #392
• Fix link in Chronos-Bolt tutorial notebook entry by @shchur in #396
• Optimize convert_df_input_to_list_of_dicts_input and validate_df_inputs by @shchur in #395
• Chronos-2: Add LoRA fine-tuning support by @abdulfatir in #393
• Chronos-2: Add option to skip dataframe validation in predict_df by @abdulfatir in #400

Full Changelog: v2.1.0...v2.2.0rc1

2.1.0

🚀 What's New?

• Chronos-2 can now be deployed on AWS via SageMaker JumpStart. Please check this notebook for details.
• Scaled Dot Product Attention (SDPA) is now used as the default attention implementation in Chronos-2. If you need to use the previous eager implementation, please load the model with Chronos2Pipeline.from_pretrained(..., attn_implementation="eager").
• predict_df support has been added for older Chronos and Chronos-Bolt models. Now, all models (Chronos-2, Chronos-Bolt, Chronos) provide a unified pandas dataframe API. Note: Only Chronos-2 supports multivariate and covariate-informed forecasting.
• Chronos2Pipeline.embed has been added, enabling users to extract embeddings from the last layer of the Chronos-2 encoder.

🐛 Bug Fixes

• Fixed issues related to past-only covariates use during Chronos-2 fine-tuning. If you're fine-tuning Chronos-2 models, we strongly recommend upgrading to chronos-forecasting==2.1.0.
• Fixed issue related to multiple workers on windows.

All Changes

• Fix pandas install instruction by @abdulfatir in #326
• Scope down GitHub token permissions by @AdnaneKhan in #328
• Update Chronos-2 notebook install instructions by @abdulfatir in #333
• Add logging steps in fine-tuning example by @abdulfatir in #334
• [chronos-2] add support for SDPA by @kashif in #331
• Treat notebooks as documentation in linguist by @abdulfatir in #338
• Count notebooks as python in language stats by @abdulfatir in #347
• Move SageMaker notebook to a new path by @shchur in #348
• Revert "Move SageMaker notebook to a new path" by @shchur in #349
• fix validate_and_prepare_single_dict_task by separating "past only" and "future known" cov keys by @HarvestStars in #344
• Add Chronos-2 SageMaker notebook by @shchur in #350
• Update model link to Hugging Face by @shchur in #353
• Remove logo from README by @shchur in #355
• Update Notebooks and README by @abdulfatir in #356
• Add FutureWarning for CloudFront by @abdulfatir in #357
• Relax transformers lower bound to >=4.41 by @abdulfatir in #364
• Set num_workers=0 in Chronos-2 test data loader by @IliasAarab in #365
• Update long horizon heuristic for Chronos-Bolt by @abdulfatir in #366
• Add predict_df support for Chronos and Chronos-Bolt by @abdulfatir in #371
• Add Chronos2Pipeline.embed by @abdulfatir in #361
• Mask past-only covariates during loss computation by @abdulfatir in #379
• Chronos-2: Change default fine-tuning learning rate and remove experimental label by @abdulfatir in #381
• Update version to 2.1.0rc1 for pre-release by @abdulfatir in #383
• Update torch dependency version to >=2.2,<3 by @abdulfatir in #388
• Bump version to 2.1.0 by @abdulfatir in #389

New Contributors

• @AdnaneKhan made their first contribution in #328
• @kashif made their first contribution in #331
• @HarvestStars made their first contribution in #344
• @IliasAarab made their first contribution in #365

Full Changelog: v2.0.1...v2.1.0

2.1.0rc1

What's Changed

• Fix pandas install instruction by @abdulfatir in #326
• Scope down GitHub token permissions by @AdnaneKhan in #328
• Update Chronos-2 notebook install instructions by @abdulfatir in #333
• Add logging steps in fine-tuning example by @abdulfatir in #334
• [chronos-2] add support for SDPA by @kashif in #331
• Treat notebooks as documentation in linguist by @abdulfatir in #338
• Count notebooks as python in language stats by @abdulfatir in #347
• Move SageMaker notebook to a new path by @shchur in #348
• Revert "Move SageMaker notebook to a new path" by @shchur in #349
• fix validate_and_prepare_single_dict_task by separating "past only" and "future known" cov keys by @HarvestStars in #344
• Add Chronos-2 SageMaker notebook by @shchur in #350
• Update model link to Hugging Face by @shchur in #353
• Remove logo from README by @shchur in #355
• Update Notebooks and README by @abdulfatir in #356
• Add FutureWarning for CloudFront by @abdulfatir in #357
• Relax transformers lower bound to >=4.41 by @abdulfatir in #364
• Set num_workers=0 in Chronos-2 test data loader by @IliasAarab in #365
• Update long horizon heuristic for Chronos-Bolt by @abdulfatir in #366
• Add predict_df support for Chronos and Chronos-Bolt by @abdulfatir in #371
• Add Chronos2Pipeline.embed by @abdulfatir in #361
• Mask past-only covariates during loss computation by @abdulfatir in #379
• Chronos-2: Change default fine-tuning learning rate and remove experimental label by @abdulfatir in #381
• Update version to 2.1.0rc1 for pre-release by @abdulfatir in #383

New Contributors

• @AdnaneKhan made their first contribution in #328
• @kashif made their first contribution in #331
• @HarvestStars made their first contribution in #344
• @IliasAarab made their first contribution in #365

Full Changelog: v2.0.1...v2.1.0rc1

v2.0.1

What's Changed

• Relax transformers lower bound to >=4.41 by @abdulfatir in #364
• Update long horizon heuristic for Chronos-Bolt by @abdulfatir in #366

Full Changelog: v2.0.0...v2.0.1