README.md

---
title: Chronos-2 Zero-Shot Demo
emoji: 📈
colorFrom: blue
colorTo: red
sdk: docker
app_port: 7860
---

# Chronos-2 Zero-Shot Forecasting Demo

Interactive demo showcasing Amazon Chronos-2's zero-shot time series forecasting capability.

## Features

- **Zero-Shot Forecasting**: No training or fine-tuning required
- **Real-time Prediction**: Input data and get instant forecasts
- **Probabilistic Output**: Median prediction with 80% confidence interval
- **Comparison Tab**: Visual comparison of traditional ML vs Chronos pipeline

## Usage

1. Enter time series values (comma or newline separated)
2. Set prediction steps (1-24)
3. Get instant forecast with confidence intervals

## Model

- **Chronos-2** (amazon/chronos-2): 120M parameter encoder-only foundation model
- Supports context length up to 8192 tokens
- CPU inference (GPU optional)

## Local Development

```bash
pip install -r requirements.txt
uvicorn server:app --reload --port 7860
```

## References

- [Chronos-2 GitHub](https://github.com/amazon-science/chronos-forecasting)
- [Chronos-2 Paper](https://arxiv.org/abs/2410.03847)