metadata
pipeline_tag: time-series-forecasting
tags:
- time series
- time series foundation models
- time series forecasting
- zero-shot
LightGTS: A Lightweight General Time Series Forecasting Model
🚩 News (2025.06) LightGTS has been accepted as ICML 2025.
Introduction
Quick Demos
pip install transformers==4.30.2 # Use this version for stable compatibility
import torch
from transformers import AutoModelForCausalLM
# load pretrain model
# supports different lookback/forecast lengths
model = AutoModelForCausalLM.from_pretrained('DecisionIntelligence/LightGTS', trust_remote_code=True)
# prepare input
batch_size, lookback_length = 1, 528
seqs = torch.randn(batch_size, lookback_length).unsqueeze(-1).float()
# Note that Sundial can generate multiple probable predictions
forecast_length = 192
outputs = model.generate(seqs, patch_len = 48, stride_len=48, max_output_length=forecast_length, inference_patch_len=48)
print(output.shape)
Citation
If you find Sundial helpful for your research, please cite our paper:
@article{wang2025lightgts,
title={LightGTS: A Lightweight General Time Series Forecasting Model},
author={Wang, Yihang and Qiu, Yuying and Chen, Peng and Shu, Yang and Rao, Zhongwen and Pan, Lujia and Yang, Bin and Guo, Chenjuan},
journal={arXiv preprint arXiv:2506.06005},
year={2025}
}