NanoForecast v0.3

6.5M-Parameter Time Series Foundation Model — Deploy Anywhere

CPU inference · Raspberry Pi · ONNX · Streaming · Quantile forecasts

Built by Eulogik — deployable AI for the real world

What is NanoForecast v0.3?

NanoForecast v0.3 is the initial release of a 6.5M-parameter time series foundation model designed for deployment on CPUs, Raspberry Pi, edge devices, and in the browser. It performs zero-shot forecasting with quantile uncertainty bounds.

This is a reference release. For the latest improvements, see eulogik/nanoforecast-v05 (43.8% better MASE with same architecture).

Key Features

Zero-shot forecasting — no training needed for new time series
Streaming inference — feed one value at a time via stateful DeltaNet RNN
Quantile predictions — p10, p25, p50, p75, p90 with monotonic guarantees
ONNX export — 9.2MB INT8 for edge, IoT, browser deployment
CPU inference — 19.5ms median latency on Apple M4
Train from CSV — fine-tune on your data in minutes
Apache 2.0 license — no restrictions on commercial use

Benchmark Results

Standard protocol: context 512, horizon 48, non-overlapping test windows, MASE scaled by seasonal-naive in-sample MAE.

Dataset	NanoForecast v0.3 (6.5M)	TimesFM (200M)	PatchTST (15M+)
ETTh1	0.681	0.705	0.781
ETTh2	1.328	1.360	1.467
ETTm1	0.288	0.545	0.488
exchange_rate	11.758	4.383	3.861
electricity	2.213	0.923	1.347
traffic	1.913	0.765	1.379
Overall MASE	3.030	1.447	1.554

v0.3 → v0.5 improvement: Overall MASE 3.030 → 1.704 (↓43.8%) with same architecture, pipeline fixes only.

Quick Start

Install

pip install nanoforecast

Zero-Shot Forecasting

import numpy as np
from nanoforecast import NanoForecast

model = NanoForecast.from_pretrained("eulogik/nanoforecast-v03")

context = np.sin(np.linspace(0, 8*np.pi, 512)) + 0.1 * np.random.randn(512)
result = model.predict(context, horizon=48, freq=1)

print(result["forecast"].shape)     # (48,) point forecast
print(result["quantiles"].shape)    # (5, 48)  p10..p90

Streaming Inference

result = model.predict(context, horizon=48, return_state=True)
state = result.pop("state")

for new_val in incoming_stream:
    result = model.predict_step(new_val, state, horizon=48)
    forecast = result["forecast"][0]

Model Files

File	Size
`model.safetensors`	26.1 MB
`config.json`	343 B
`model_card.json`	710 B

Citation

@article{kishore2026nanoforecast,
  title={NanoForecast v0.5: Competitive Time Series Forecasting Through Training Pipeline Optimization},
  author={Kishore, Gautam},
  journal={arXiv preprint arXiv:2609.31669},
  year={2026}
}

Paper: arxiv.org/abs/2609.31669

eulogik/nanoforecast-v03