nanoforecast-500k / README.md

GautamKishore

SEO-optimized model card with cross-links, badges, viral copy

98e273a verified 2 months ago

preview code

Raw

History Blame Contribute Delete

5.03 kB

metadata

license: apache-2.0
library_name: pytorch
pipeline_tag: time-series-forecasting
tags:
  - time-series
  - forecasting
  - pytorch
  - deployable
  - edge-ai
  - onnx
  - streaming
  - raspberry-pi
  - transformer
  - zero-shot
  - iot
  - real-time
  - tiny-ml
  - timesfm-alternative
  - huggingface
metrics:
  - mase
  - smape
  - mae
  - crps
model-index:
  - name: NanoForecast 500k
    results:
      - task:
          type: time-series-forecasting
          name: Time Series Forecasting
        dataset:
          name: ETTh1
          type: ett
          config: h1
        metrics:
          - type: mase
            value: 3.342
            name: MASE
          - type: smape
            value: 25.13
            name: sMAPE (%)
      - task:
          type: time-series-forecasting
          name: Time Series Forecasting
        dataset:
          name: ETTh2
          type: ett
          config: h2
        metrics:
          - type: mase
            value: 3.707
            name: MASE
          - type: smape
            value: 17.65
            name: sMAPE (%)
      - task:
          type: time-series-forecasting
          name: Time Series Forecasting
        dataset:
          name: ETTm1
          type: ett
          config: m1
        metrics:
          - type: mase
            value: 3.578
            name: MASE
          - type: smape
            value: 17.22
            name: sMAPE (%)
      - task:
          type: time-series-forecasting
          name: Time Series Forecasting
        dataset:
          name: Overall
          type: multi-dataset
        metrics:
          - type: mase
            value: 3.453
            name: Overall MASE
          - type: smape
            value: 18.68
            name: Overall sMAPE (%)

🔮 NanoForecast 500k (v0.2)
_{Ultra-lightweight time series transformer}

1.6M params · 256 context · Streaming RNN · ONNX-ready
Runs on CPU, Raspberry Pi, and in the browser

🚀 New: Try the improved nanoforecast-v03 — 21% better MASE, 6.5M params, 512 context!

📦 Quick Start

pip install nanoforecast

from nanoforecast import NanoForecast
model = NanoForecast.from_pretrained("eulogik/nanoforecast-500k")

🏆 Benchmarks

Dataset	MASE	sMAPE (%)	MAE	CRPS
ETTh1	3.342	25.13	2.402	1.800
ETTh2	3.707	17.65	3.212	2.518
ETTm1	3.578	17.22	1.174	1.003
exchange_rate	7.306	1.63	0.010	0.009
electricity	1.536	5.65	189.748	187.256
traffic	1.246	44.80	0.006	0.005
Overall	3.453	18.68	32.759	32.099

🔥 For better accuracy, upgrade to nanoforecast-v03 (MASE 2.73, 21% improvement).

🔄 Streaming Inference (Unique to NanoForecast)

result = model.predict(context, horizon=48, return_state=True)
state = result.pop("state")
for new_val in incoming_data_stream:
    result = model.predict_step(new_val, state, horizon=48)
    print(result["forecast"][0, :5])

Perfect for IoT, real-time dashboards, and live financial data.

📋 Model Details

Attribute	Value
Profile	`d64-L8`
Parameters	1,606,232
Context	256
Horizon	48
Size	6.4 MB (FP32), ~1.4 MB (ONNX)
Architecture	LongConv + DeltaNet RNN + Gated Router + MLP
Deploy targets	CPU, ARM, Raspberry Pi, Lambda, iOS, browser

🎯 Try It in 1 Click

Upload a CSV → forecast + prediction intervals. No code. No GPU.

📚 All Variants

Model	Params	Context	MASE	Best For
nanoforecast-200k	676K	256	~4-11	Extreme edge / RPi Zero
nanoforecast-500k (you are here)	1.6M	256	3.45	General purpose
nanoforecast-v03	6.5M	512	2.73	Max accuracy

⚡ Deploy

# FastAPI
pip install nanoforecast fastapi uvicorn python-multipart
python3 deploy/fastapi_server.py

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

❤️ Built by Eulogik

Eulogik — deployable AI for the real world.

Star the repo ⭐ on GitHub!

🔮 NanoForecast 500k (v0.2) Ultra-lightweight time series transformer