# OpenLTM

**Repository Path**: limoncc/OpenLTM

## Basic Information

- **Project Name**: OpenLTM
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: MIT
- **Default Branch**: main
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2026-01-05
- **Last Updated**: 2026-01-05

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# OpenLTM

OpenLTM is an open codebase aiming to provide a pipeline to develop and evaluate large time-series models.

> For deep time series models, we recommend [Time-Series-Library](https://github.com/thuml/Time-Series-Library) and this comprehensive [Survey](https://arxiv.org/abs/2407.13278).

> Try out our out-of-the-box time series foundation models in [Large-Time-Series-Model](https://github.com/thuml/Large-Time-Series-Model) or [HuggingFace](https://huggingface.co/collections/thuml/time-series-foundation-models-67c80ace73299239b651d954).

## Updates


:triangular_flag_on_post: **News** (2025.05) We release 260B pre-trained [Timer in PyTorch](https://cloud.tsinghua.edu.cn/f/01c35ca13f474176be7b/). This version is equivalent to [Huggingface Model](https://huggingface.co/thuml/timer-base-84m), but more user-friendly for fine-tuning. This [notebook](./load_pth_ckpt.ipynb) shows how to load and use the checkpoint.

:triangular_flag_on_post: **News** (2025.04) Many thanks for the implementation of [TTMs](https://arxiv.org/pdf/2401.03955) and other LLM4TS methods from [frndtls](https://github.com/frndtls).

:triangular_flag_on_post: **News** (2024.12) Many thanks for the implementation of [GPT4TS](https://arxiv.org/abs/2302.11939) from [khairulislam](https://github.com/khairulislam).

:triangular_flag_on_post: **News** (2024.10) We include several large time-series models, release pre-training code, and provide scripts.

## What is LTM?

LTM (**L**arge **T**ime-Series **M**odel, aka Time Series Foundation Model, TSFM) is a deep time series models built on scalable backbones (e.g., Transformers) via large-scale pre-training, which will be applied to a variety of time series data (e.g., zero-shot forecasting) and downstream tasks (e.g., general feature extraction of time series). For more information, here we list some tutorials: [[CN]](https://cloud.tsinghua.edu.cn/f/7b88e05d38bb40a1be30/), [[EN]](https://cloud.tsinghua.edu.cn/f/c5fca76d6fa54f1d891a/).

<p align="center">
<img src="./figures/overview.png" alt="" align=center />
</p>

## Model Checklist

- [x] **Timer-XL** - Timer-XL: Long-Context Transformer for Unified Time Series Forecasting. [[ICLR 2025]](https://arxiv.org/abs/2410.04803), [[Code]](https://github.com/thuml/Timer-XL)
- [x] **Moirai** - Unified Training of Universal Time Series Forecasting Transformers. [[ICML 2024]](https://arxiv.org/abs/2402.02592), [[Code]](https://github.com/SalesforceAIResearch/uni2ts)
- [x] **Timer** - Timer: Generative Pre-trained Transformers Are Large Time Series Models. [[ICML 2024]](https://arxiv.org/abs/2402.02368), [[Code]](https://github.com/thuml/Large-Time-Series-Model)
- [x] **Moment** - MOMENT: A Family of Open Time-series Foundation Model. [[ICML 2024]](https://arxiv.org/abs/2402.03885), [[Code]](https://github.com/moment-timeseries-foundation-model/moment)
- [x] **TTMs** - Tiny Time Mixers (TTMs): Fast Pre-trained Models for Enhanced Zero/Few-Shot Forecasting of Multivariate Time Series. [[Arxiv 2024]](https://arxiv.org/pdf/2401.03955), [[Code]](https://huggingface.co/ibm-research/ttm-research-r2)
- [x] **GPT4TS** - One Fits All: Power General Time Series Analysis by Pretrained LM. [[NeurIPS 2023]](https://arxiv.org/abs/2302.11939), [[Code]](https://github.com/DAMO-DI-ML/NeurIPS2023-One-Fits-All)
- [x] **Time-LLM**: . Time-LLM: Time Series Forecasting by Reprogramming Large Language Models. [[ICLR 2024]](https://arxiv.org/abs/2310.01728), [[Code]](https://github.com/KimMeen/Time-LLM)
- [x] **AutoTimes**: Autoregressive Time Series Forecasters via Large Language Models. [[NeurIPS 2024]](https://arxiv.org/abs/2402.02370), [[Code]](https://github.com/thuml/AutoTimes)
- [ ] Sundial: A Family of Highly Capable Time Series Foundation Models. [[ICML 2025]](https://arxiv.org/abs/2502.00816), [[Code]](https://github.com/thuml/Sundial)
- [ ] LLMTime: Large Language Models Are Zero-Shot Time Series Forecasters. [[NeurIPS 2023]](https://arxiv.org/abs/2310.07820), [[Code]](https://github.com/ngruver/llmtime)
- [ ] Chronos: Learning the Language of Time Series. [[TMLR 2024]](https://arxiv.org/abs/2403.07815), [[Code]](https://github.com/amazon-science/chronos-forecasting)
- [ ] Time-MoE: Billion-Scale Time Series Foundation Models With Mixture Of Experts. [[ICLR 2025]](https://arxiv.org/abs/2409.16040), [[Code]](https://github.com/Time-MoE/Time-MoE)
- [ ] A Decoder-Only Foundation Model for Time-Series Forecasting. [[ICML 2024]](https://arxiv.org/abs/2310.10688), [[Code]](https://github.com/google-research/timesfm)


## Usage

1. Install Python 3.11 For convenience, execute the following command.

```
pip install -r requirements.txt
```

2. Place downloaded data in the folder ```./dataset```. Here is a [dataset summary](./figures/datasets.png).

- For univariate pre-training (skip this step if you use a pre-trained checkpoint):
  * [UTSD](https://huggingface.co/datasets/thuml/UTSD) contains 1 billion time points for large-scale pre-training (in numpy format): [[Download]](https://cloud.tsinghua.edu.cn/f/93868e3a9fb144fe9719/
).
  * [ERA5-Family](https://www.ecmwf.int/en/forecasts/dataset/ecmwf-reanalysis-v5) (40-year span, thousands of variables) for domain-specific model: [[Download]](https://cloud.tsinghua.edu.cn/f/7fe0b95032c64d39bc4a/).

- For supervised training or modeling adaptation
  * Datasets from [TSLib](https://github.com/thuml/Time-Series-Library) : [[Download]](https://cloud.tsinghua.edu.cn/f/4d83223ad71047e28aec/).

3. Training from scratch or using our pre-trained checkpoint in the folder ```./checkpoints```

- We provide the [checkpoint](https://arxiv.org/abs/2410.04803) pre-trained on 260B time points [[Download]](https://cloud.tsinghua.edu.cn/f/01c35ca13f474176be7b/).

4. We provide some supervised training, pre-training, and adaptation scripts under the folder `./scripts/`:

```
# Supervised training
# (a) one-for-one forecasting
bash ./scripts/supervised/forecast/moirai_ecl.sh
# (b) one-for-all (rolling) forecasting
bash ./scripts/supervised/rolling_forecast/timer_xl_ecl.sh

# Large-scale pre-training
# (a) pre-training on UTSD
bash ./scripts/pretrain/timer_xl_utsd.sh
# (b) pre-training on ERA5
bash ./scripts/pretrain/timer_xl_era5.sh

# Model adaptation
# (a) full-shot fine-tune
bash ./scripts/adaptation/full_shot/timer_xl_etth1.sh
# (b) few-shot fine-tune
bash ./scripts/adaptation/few_shot/timer_xl_etth1.sh
```

5. Develop your large time-series model.

- Add the model file to the folder `./models`. You can follow the `./models/timer_xl.py`.
- Include the newly added model in the `Exp_Basic.model_dict` of  `./exp/exp_basic.py`.
- Create the corresponding scripts under the folder `./scripts`.

## Recommended Resources

Here we list some resources of LTMs, which support out-of-box usage (e.g., zero-shot forecasting):

- Sundial: https://huggingface.co/thuml/sundial-base-128m
- Timer: https://huggingface.co/thuml/timer-base-84m
- Chronos: https://huggingface.co/amazon/chronos-t5-base
- Moirai: https://huggingface.co/Salesforce/moirai-1.0-R-base
- TimesFM: https://huggingface.co/google/timesfm-1.0-200m
- Time-MoE: https://huggingface.co/Maple728/TimeMoE-50M
- TTMs: https://huggingface.co/ibm-research/ttm-research-r2

> [!NOTE]
> LTMs are still small compared to foundation models of other modalities. For example, it is okay to use CPUs for inference, RTX 4090s for adaptation, A100s for pre-training.

## Citation

If you find this repo helpful, please cite our paper. 

```
@inproceedings{liutimer,
  title={Timer: Generative Pre-trained Transformers Are Large Time Series Models},
  author={Liu, Yong and Zhang, Haoran and Li, Chenyu and Huang, Xiangdong and Wang, Jianmin and Long, Mingsheng},
  booktitle={Forty-first International Conference on Machine Learning}
}

@article{liu2024timer,
  title={Timer-XL: Long-Context Transformers for Unified Time Series Forecasting},
  author={Liu, Yong and Qin, Guo and Huang, Xiangdong and Wang, Jianmin and Long, Mingsheng},
  journal={arXiv preprint arXiv:2410.04803},
  year={2024}
}

@article{liu2025sundial,
  title={Sundial: A Family of Highly Capable Time Series Foundation Models},
  author={Liu, Yong and Qin, Guo and Shi, Zhiyuan and Chen, Zhi and Yang, Caiyin and Huang, Xiangdong and Wang, Jianmin and Long, Mingsheng},
  journal={arXiv preprint arXiv:2502.00816},
  year={2025}
}
```

## Acknowledgment

We appreciate the following GitHub repos a lot for their valuable code and efforts:
- Time-Series-Library (https://github.com/thuml/Time-Series-Library)
- Large-Time-Series-Model (https://github.com/thuml/Large-Time-Series-Model)
- AutoTimes (https://github.com/thuml/AutoTimes)

## Contributors

If you have any questions or want to use the code, feel free to contact:
* Yong Liu (liuyong21@mails.tsinghua.edu.cn)
* Guo Qin (qinguo24@mails.tsinghua.edu.cn)
* Haixuan Liu (liuhaixu21@mails.tsinghua.edu.cn)