Falcon-TST: A Large-Scale Time Series Foundation Model

A large-scale time series foundation model utilizing Mixture of Experts (MoE) architecture with multiple patch tokenizers for efficient and accurate time series forecasting.

📖 Introduction

Falcon-TST is a cutting-edge time series foundation model that leverages the power of Mixture of Experts (MoE) architecture combined with multiple patch tokenizers. This innovative approach enables efficient processing of time series data while maintaining high accuracy across various forecasting tasks.

You can find more details about the model on GitHub page.

🚀 Quick Start

import torch
from transformers import AutoModel

# Load pre-trained model (when available)
model = AutoModel.from_pretrained(
    'ant-intl/Falcon-TST_Large',
    trust_remote_code=True
)

# Prepare your time series data
batch_size, lookback_length, channels = 1, 2880, 7
time_series = torch.randn(batch_size, lookback_length, channels)

# Load the model and data to the same device
device = torch.cuda.current_device() if torch.cuda.is_available() else 'cpu'
model = model.to(device)
time_series = time_series.to(device)

# Generate forecasts
forecast_length = 96
predictions = model.predict(time_series, forecast_horizon=forecast_length)