Practical overview
When it fits
Use TEMPO when you want to investigate prompt-based pretrained forecasting or build a Python prototype that turns a sequence of historical values into a requested forecast horizon.
Who it is for
- Researchers exploring pretrained and prompt-based forecasting methods
- Developers prototyping time-series forecasting workflows
- Users who want to reproduce the repository’s zero-shot experiments or try its pretrained model
Documented uses
- Forecast future values from historical time-series inputs.
- Prototype an inference workflow using pretrained forecasting weights.
- Reproduce documented zero-shot experiments on ETTh2 or a customer dataset.
- Explore repository scripts and data for pretraining and inference experiments.
Requirements & limits
The model code is publicly available on GitHub, and the repository identifies an MIT license. The README links pretrained weights on Hugging Face. Its license badge text differs from the repository’s MIT license metadata; check the license file before reuse.
- The README documents both a concise pretrained inference example and separate dataset-based experiment scripts; choose the route that matches your task.
- The example uses Python, TimeAGI, NumPy and PyTorch, and chooses CUDA when available or CPU otherwise.
- The README’s example uses a 336-value input and requests 96 predicted values; these are illustrative values rather than stated universal limits.
- The README points to pretrained weights on Hugging Face and Colab demonstrations for ETTh2 and customer-dataset inference.
- The repository metadata identifies MIT, while the README displays a license badge labeled Apache-2.0; review the repository’s license file before using the code.
Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.
Set up the documented Python environment
Create the Conda environment and install TimeAGI, as shown in the TEMPO README.
Before you start
- Conda
- Python 3.8 environment
Steps
Create a Conda environment named tempo with Python 3.8.
Activate the tempo environment.
Install the timeagi package with pip.
Expected outcome
The README’s documented environment setup is complete; it does not by itself run a forecast.
Guide limitations
- The README does not specify further dependency or hardware requirements for this setup sequence.
Load the pretrained checkpoint and request a forecast
Use the README’s Python example to pass historical values to TEMPO and request predicted values.
Before you start
- The documented Python environment with TimeAGI installed
- NumPy and PyTorch imports, as used in the README example
- Access to the pretrained model repository Melady/TEMPO and checkpoint TEMPO-80M_v1.pth
Steps
Import NumPy, PyTorch and TEMPO from tempo.models.TEMPO.
Load the pretrained model with repo_id Melady/TEMPO, filename TEMPO-80M_v1.pth and the documented checkpoint cache directory.
Provide a one-dimensional NumPy input sequence and call model.predict with a pred_length.
Print the predicted values.
Expected outcome
The example produces predicted values for the requested forecast length; its illustration uses an input of 336 random values and pred_length 96.
Guide limitations
- The input in the README is random demonstration data, not a real-world dataset.
- The README selects CUDA when available and CPU otherwise; it does not establish broader platform compatibility.
Sources & review
Source record details
- Source availability
- Active
- Sources checked
- 16 October 2026
- Original website title
- DC-research/TEMPO · Official homepage title · retained in its source language