nanoforecast-200k

NanoForecast is the world's most deployable time series forecasting model (~676K parameters). It trains on a laptop, runs on a Raspberry Pi, and exports to 1.4 MB ONNX for edge/IoT/browser deployment.

This is the v0.1 checkpoint — the smallest variant, designed for ultra-constrained environments (Raspberry Pi Zero, browser, Lambda). For better accuracy, use nanoforecast-500k (1.6M params, 6-dataset training).

Built by Eulogik — deployable AI for the real world.

Model details

• Profile: nano-200k
• Parameters: 676,108
• Context length: 256
• Prediction length: 48
• Patch size: 8
• Hidden dim / layers: 32 / 4
• Quantiles: [0.1, 0.25, 0.5, 0.75, 0.9]
• Architecture: LongConv + DeltaNet RNN + Gated Router + MLP
• FP32 size: ~2.7 MB
• INT8 size: ~0.7 MB

Training

• Dataset: ETTh1 (hourly temperature)
• Epochs: 20
• Best epoch: 9 (val_loss 11.28)
• Learning rate: 1e-4
• Batch size: 32

When to use this checkpoint

Use nanoforecast-200k when you need the absolute smallest model:

• Raspberry Pi Zero / Pico deployments
• AWS Lambda (cold start < 100ms)
• Browser-based ONNX.js (sub-1 MB download)
• Battery-powered IoT sensors

Use nanoforecast-500k when you need better accuracy and can spare 6.4 MB:

• Standard Raspberry Pi 4/5 deployments
• Docker / FastAPI server
• Real-time streaming with DeltaNet stateful inference

Quickstart

import numpy as np
from nanoforecast import NanoForecast

model = NanoForecast.from_pretrained('eulogik/nanoforecast-200k')
context = np.sin(np.linspace(0, 8*np.pi, 256)) + 0.1 * np.random.randn(256)
out = model.predict(context, horizon=48, freq=1)
print(out['forecast'].shape)  # (48,) point forecast

Deploy

# ONNX export (0.7 MB INT8)
pip install "nanoforecast[onnx]"
python3 -m nanoforecast.export.onnx_export --checkpoint <checkpoint-dir> --output nanoforecast.onnx

Known limitations

This v0.1 checkpoint was trained on a single dataset (ETTh1, 20 epochs). Accuracy is limited (MASE ~4-11 on ETT benchmarks). What it does well: being the smallest deployable TS model on the Hub. Train on your own data for better accuracy.

Attribution

Built by Eulogik — deployable AI for the real world.