v0.2.0 - Fine-tuning & Exogenous Variables

What's New

Fine-tuning Support

• FinetuneDataModule: PyTorch Lightning DataModule for training on custom HuggingFace datasets
• TotoForFinetuning: Lightning module with default configuration matching Toto's pretraining recipe
• Ready-to-use training script (finetune_toto.py) with YAML configuration
• Interactive tutorial notebook (finetuning_tutorial.ipynb)

Exogenous Variable Support

• Extended TotoBackbone to handle known future covariates (e.g., weather forecasts, scheduled events)
• Fusion module for combining target and exogenous variate embeddings with learnable labels
• Updated TotoForecaster to inject future exogenous values during autoregressive decoding

Evaluation & Benchmarking

• Custom evaluation on FEV datasets that are not included in Toto's pretraining corpus
• Evaluation scripts comparing zero-shot, fine-tuned, and fine-tuned with exogenous approaches
• Pre-computed results on 14 datasets (ENTSOE, EPF, Solar, UCI Air Quality, Rohlik)

Installation

pip install toto-ts==0.2.0

Full Changelog: v0.1.4...v0.2.0

0.1.4

Adds logging when the xformers library is available.

0.1.3

Upgrade github actions

0.1.2

Upgrading github actions hash for publishing to pypi

0.1.1

Patch release for toto to publish to pypi

0.1.0

Initial release for Toto: Timeseries Optimized Transformer for Observability