shinydatatech/forecastagent-v1.0

ForecastAgent 1.0

ForecastAgent 1.0 is a zero-shot and fine-tuned time series forecasting foundation model powered by an xLSTM backbone architecture.

Model Details

• Backbone: xLSTM (based on temporal/variate mixers)
• Parameters: ~82.5M
• Task: Zero-Shot Probabilistic Time Series Forecasting
• Output: 9 Quantiles (10% to 90% percentiles)
• License: Apache-2.0

Usage

You can load and use this model directly using the forecast-agent-sdk Python SDK:

from forecastagent import ForecastAgent

# Load the model from Hugging Face
agent = ForecastAgent.from_pretrained("shinydatatech/forecastagent-v1.0")

# Input target history (e.g., hourly electricity consumption)
history = [10.2, 11.5, 12.1, 11.8, 13.0, 14.5, 15.2, 14.8, 13.9, 13.1]

# Predict 24 steps forward
results = agent.predict(
    target=history,
    prediction_length=24,
    freq="h"
)

print("Median Forecast:", results["median"])
print("10th percentile:", results["lower"])
print("90th percentile:", results["upper"])

Developers & Licensing

Developed by ShinyDataTech (shinydatatech@gmail.com). Licensed under Apache 2.0.