Fine-tune small time-series foundation models on your own data. Rough edges included.
Status: early-stage, under active development. Not production-ready. Expect breaking changes.
DeepTempo is a toolkit for fine-tuning pre-trained time-series transformer models on domain-specific forecasting tasks. The goal is to lower the barrier for practitioners who want to adapt foundation models to their own data without writing everything from scratch.
Think of it as "HuggingFace Trainer meets time-series" — a narrow, focused tool, not a platform.
- A model zoo with hundreds of pre-trained weights
- A SaaS or managed service
- Production-hardened infrastructure
- A research framework for designing new architectures