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

Uni 3Dar

Uni 3Dar is an open-source Python research implementation of an autoregressive model for 3D generation and understanding, focused on structures such as molecules and crystals. Its repository provides pretrained checkpoints, datasets and scripts for molecule generation, crystal generation and prediction, alongside training and molecular-property fine-tuning workflows.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

Uni 3Dar — 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 Uni 3Dar when you want research code and documented command-line workflows for pretrained or train-from-scratch molecular and crystal experiments, rather than a hosted, point-and-click service.

Who it is for

  • Researchers reproducing the Uni-3DAR paper’s molecule and crystal experiments.
  • Machine-learning practitioners exploring autoregressive methods for 3D structure generation and understanding.
  • Researchers fine-tuning released representations for molecular-property prediction.

Documented uses

  • Generate molecules using pretrained QM9 or DRUG checkpoints.
  • Generate MP20 crystals, predict crystal structures, or run PXRD-guided crystal structure prediction.
  • Train the documented QM9, DRUG and MP20 pipelines from scratch.
  • Fine-tune pretrained representations for documented molecular-property tasks such as HOMO, LUMO and GAP.
Before you choose

Requirements & limits

The public repository declares an MIT license. The README identifies Uni-Core as a dependency and provides a Docker image as a convenience.

  • The implementation is Python and depends on Uni-Core; the repository lists a prebuilt Docker image.
  • Pretrained checkpoints and dataset archives are downloaded from the project’s Hugging Face repository.
  • Training defaults vary by workflow: four GPUs for QM9 and de-novo MP20 generation, and eight GPUs for DRUG and MP20 crystal-prediction variants.
  • The README documents research scripts and local data workflows, not a hosted interface or API.
  • Claims about speed and generation quality are reported by the project itself.

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

Generate QM9 molecules with a pretrained Uni-3DAR model

Run the documented QM9 inference script using its pretrained checkpoint.

Before you start

  • Uni-Core dependency; the README also provides a prebuilt Docker image as a convenience.
  • The qm9.pt pretrained checkpoint, downloadable from the Uni-3DAR Hugging Face repository.

Steps

  1. Download qm9.pt from the [Uni-3DAR Hugging Face repository](https://huggingface.co/dptech/Uni-3DAR/tree/main).

  2. From the repository, run the command below.

    Documented command
    bash scripts/inference_qm9.sh qm9.pt

Expected outcome

The inference script is intended to generate QM9 molecules using the pretrained model.

Guide limitations

  • The README does not specify runtime requirements or describe the output format.
  • The README identifies Uni-Core as a dependency but does not provide setup steps for it.
Documentation 1 ↗
Source-based practical guide

Generate MP20 crystals with a pretrained Uni-3DAR model

Run the documented MP20 de-novo crystal generation inference script.

Before you start

  • Uni-Core dependency; the README also provides a prebuilt Docker image as a convenience.
  • The mp20.pt checkpoint and mp20_data.tar.gz dataset archive, downloadable from the Uni-3DAR Hugging Face repository.

Steps

  1. Download mp20.pt and mp20_data.tar.gz from the [Uni-3DAR Hugging Face repository](https://huggingface.co/dptech/Uni-3DAR/tree/main).

  2. Extract the dataset with tar -xzvf mp20_data.tar.gz.

  3. From the repository, run the command below.

    Documented command
    bash scripts/inference_mp20.sh mp20.pt ./mp20_data/test.csv

Expected outcome

The inference script is intended to perform de-novo MP20 crystal generation using the pretrained checkpoint and test CSV.

Guide limitations

  • The README does not specify runtime requirements or describe the output format.
  • The README identifies Uni-Core as a dependency but does not provide setup steps for it.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
17 October 2026
Original website title
dptech-corp/Uni-3DAR · Official homepage title · retained in its source language
How we review sourcesSuggest a correction