🔮 NanoForecast — Deployable Time Series Forecasting

Upload your CSV or use an example. The smallest deployable time series model on the Hub (~1.6M params, runs on a Raspberry Pi, exports to 1.4 MB ONNX).

GitHub · Model on HF · Built by Eulogik

64 512
12 192
Data frequency

Forecast Table

📥 Download options

Once you have a forecast, you can export the model to ONNX for production:

python3 -m nanoforecast.export.onnx_export \
    --checkpoint checkpoints/nanoforecast-onnx \
    --output nanoforecast.onnx

Or deploy instantly with our FastAPI server.

âš¡ Why NanoForecast?

  • Tiny: 200K–1.6M params, 1.4 MB INT8 quantized
  • Fast: <50ms inference on CPU, 12ms on edge hardware
  • Deployable: ONNX → browser / Lambda / Raspberry Pi / iOS
  • Streaming: Stateful DeltaNet — feed one value at a time
  • Complete: Point forecast + intervals + decomposition in one pass