Skip to main content

​
Samay: Time-Series Foundation Models Library

Samay is a comprehensive package for training and evaluating time-series foundation models. It provides a unified interface to work with cutting-edge pre-trained models, enabling zero-shot forecasting, fine-tuning, and various time-series analysis tasks.

​
Key Features

  • 10+ Pre-trained Models: Access to state-of-the-art time-series foundation models
  • Zero-Shot Forecasting: Make predictions without any training on your specific dataset
  • Fine-Tuning Support: Adapt pre-trained models to your specific use cases
  • Unified API: Consistent interface across all models for easy switching and comparison
  • Multiple Tasks: Support for forecasting, classification, anomaly detection, and imputation
  • Production Ready: Tested on Python 3.11-3.13 with support for CPU, GPU (NVIDIA), and distributed training

​
Supported Models

Samay includes implementations of the following state-of-the-art time-series foundation models:

LPTM

Large Pre-trained Time Series Models for cross-domain analysis

MOMENT

Multi-task model for forecasting, classification, and anomaly detection

TimesFM

Google’s Time Series Foundation Model with 200M parameters

Chronos

Amazon’s language-model-based probabilistic forecasting

MOIRAI

Salesforce’s unified time-series forecasting model

TinyTimeMixers

Lightweight and efficient time-series mixing architecture

TimeMoE

Mixture of Experts architecture for time-series forecasting

Chronos 2.0

Next generation of Amazon’s Chronos with improved performance

Installation

Get Samay installed in your environment

Quick Start

Start forecasting in 5 minutes

API Reference

Detailed API documentation

​
Use Cases

  • Business Forecasting: Sales, demand, revenue predictions
  • IoT & Sensor Data: Equipment monitoring and predictive maintenance
  • Financial Analysis: Stock prices, market trends
  • Energy & Utilities: Load forecasting, consumption patterns
  • Healthcare: Patient monitoring, epidemic forecasting
  • Weather & Climate: Temperature, precipitation predictions

​
System Requirements

  • Python: 3.11, 3.12, or 3.13
  • Operating Systems: Linux, MacOS (Windows support planned)
  • Hardware: CPU, NVIDIA GPUs (Apple Silicon GPU support planned)
  • Dependencies: PyTorch, transformers, gluonts, and more

​
Research & Citations

Samay is built on cutting-edge research in time-series foundation models. If you use this library in your research, please cite: 
@inproceedings{kamarthi2024large,
  title={Large Pre-trained time series models for cross-domain Time series analysis tasks},
  author={Harshavardhan Kamarthi and B. Aditya Prakash},
  booktitle={The Thirty-eighth Annual Conference on Neural Information Processing Systems},
  year={2024},
  url={https://openreview.net/forum?id=vMMzjCr5Zj}
}

​
Community & Support

For questions, feedback, or contributions:

Build docs developers (and LLMs) love

Get started for freeTalk to us