OpenViSTAI · Agents · Blockchain · Crypto
Submit a project
Menu
AI Models

Uni2Ts

Uni2Ts is an open-source PyTorch library for research and applications in time-series forecasting. It provides a unified framework for pretraining, fine-tuning, inference and evaluation of universal time-series transformers, including workflows and pretrained resources for Moirai-family models.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Uni2Ts — OpenViST project profile in AI Models.
OpenViST project illustration. See the linked sources for product details.
Understand the project

Practical overview

When it fits

Use Uni2Ts when you want a Python-based research or development workflow for training, fine-tuning, running or evaluating time-series forecasting transformers, rather than a hosted forecasting service.

Who it is for

  • Researchers developing or studying time-series forecasting transformers.
  • Developers building forecasting workflows with pretrained models.
  • Teams that need to train, fine-tune or evaluate time-series forecasting models.

Documented uses

  • Train universal time-series forecasting transformers across datasets.
  • Prepare pretrained Moirai-family models and generate forecasts from time-series data.
  • Fine-tune forecasting models for a task or benchmark.
  • Set up rolling train/test evaluations and assess forecasting models.
Before you choose

Requirements & limits

The README documents installation from source or via PyPI. The repository specifies the Apache-2.0 license.

  • The documented inference example uses wide pandas DataFrame input converted to a GluonTS PandasDataset.
  • The example selects among Moirai, Moirai-MoE and Moirai2 model names, with configurable model size and forecasting parameters.
  • Pretrained model weights are obtained from Hugging Face.
  • Source installation includes notebook extras and creation of a .env file; the README also offers a PyPI installation command.

Check current requirements in the official documentation. Source-based guidance does not mean OpenViST has run or independently tested this project.

Source-based practical guide

Install Uni2Ts from source

Prepare the library and its notebook extras in a virtual environment.

Before you start

  • Python environment capable of running virtualenv and pip

Steps

  1. Clone the Uni2Ts repository and change into its directory.

  2. Create and activate a virtual environment.

  3. Install the repository in editable mode with the notebook extras.

  4. Create a .env file in the repository directory.

Expected outcome

The repository is installed with its notebook extras and has the .env file used in the documented setup.

Guide limitations

  • These steps cover installation only; they do not run a forecast.
Documentation 1 ↗
Source-based practical guide

Prepare a zero-shot forecasting example

Set up time-series data and a pretrained model for inference using the README workflow.

Before you start

  • Uni2Ts installed
  • A wide pandas DataFrame containing time-series data
  • A selected Moirai-family pretrained model and access to its Hugging Face weights
  • pandas and GluonTS, as used by the example

Steps

  1. Read the time-series data into a pandas DataFrame.

  2. Convert the DataFrame to a GluonTS PandasDataset.

  3. Split the dataset into train and test portions and generate rolling test instances.

  4. Select a model family and size, set forecast parameters, and prepare the model using pretrained Hugging Face weights.

Expected outcome

The example prepares a pretrained model and rolling evaluation data for a zero-shot forecasting workflow.

Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
12 October 2026
Original website title
SalesforceAIResearch/uni2ts · Official homepage title · retained in its source language
How we review sourcesSuggest a correction