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

SE3 Transformer Pytorch

SE3 Transformer Pytorch is a Python library that implements SE(3)-equivariant self-attention in PyTorch. It accepts feature and coordinate inputs and includes examples for atom tokens, molecular edges and sparse-neighbor connectivity.

Explore the practical guide ↓

Independent OpenViST project profile, based on linked sources.

SE3 Transformer Pytorch — 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 it when a PyTorch research experiment needs an SE(3)-equivariant attention component and you can prepare compatible feature, coordinate and mask inputs. For AlphaFold2-related work, note that the README recommends Equiformer instead.

Who it is for

  • Researchers experimenting with geometric deep learning on 3D data.
  • Developers prototyping molecular or protein-structure modeling components in PyTorch.
  • Practitioners exploring equivariant attention for drug-discovery research.

Documented uses

  • Apply an SE(3)-Transformer model to feature tensors paired with 3D coordinates.
  • Represent atoms as tokens or feature vectors and include bond or other edge information.
  • Constrain attention using a supplied adjacency mask for known point connectivity.
  • Explore coordinate-refinement patterns shown in the README.
Before you choose

Requirements & limits

The public GitHub repository identifies the project as MIT-licensed open source.

  • The repository provides a model component and examples, not a complete drug-discovery or protein-folding application.
  • The README documents basic feature-and-coordinate inputs, atom-token inputs, edge inputs and adjacency-based sparse neighbors.
  • The README recommends Equiformer instead and warns that versions before 0.6.0 had a bug affecting nearest-neighbor use without adjacency sparse-neighbor settings.
  • The examples use PyTorch tensors and require users to prepare the data and choose model parameters.

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 the library and prepare a basic model call

Set up the documented package import and form a basic SE3Transformer call with feature, coordinate and mask tensors.

Before you start

  • Python and PyTorch, as shown in the README examples.
  • Feature tensors, 3D coordinate tensors and a boolean mask with compatible dimensions.

Steps

  1. Install the package with the command below.

    Documented command
    pip install se3-transformer-pytorch

  2. Import torch and SE3Transformer from se3_transformer_pytorch.

  3. Configure the model with parameters such as dim, heads, depth, dim_head and num_degrees.

  4. Pass features, coordinates and a boolean mask to the model.

Expected outcome

A model call returns an output tensor; the README’s basic example uses inputs shaped (1, 1024, 512), (1, 1024, 3) and (1, 1024).

Guide limitations

  • The example uses random tensors and does not document a trained model or a complete scientific workflow.
  • The README recommends Equiformer instead and warns about a bug in earlier versions affecting nearest-neighbor use without adjacency sparse-neighbor settings.
Documentation 1 ↗
Source-based practical guide

Provide atom and bond inputs in a model call

Follow the README pattern for passing atom tokens, coordinates, a mask and discrete bond-edge tokens.

Before you start

  • Python and PyTorch.
  • Atom-token, coordinate, mask and bond-token tensors shaped consistently with the example.

Steps

  1. Install the package with the command below.

    Documented command
    pip install se3-transformer-pytorch

  2. Configure SE3Transformer with num_tokens for atom types, num_edge_tokens for edge types, and an edge_dim.

  3. Prepare atom tokens, coordinates, a boolean mask and a bond tensor.

  4. Call the model with the bond tensor passed as edges and select a return type.

Expected outcome

The example produces a model prediction from atom and bond inputs, with output shape shown as (2, 32, 1).

Guide limitations

  • The example illustrates an input pattern, not a validated molecular prediction workflow.
  • The README recommends Equiformer instead and flags the earlier-version nearest-neighbor issue.
Documentation 1 ↗
Evidence and transparency

Sources & review

Official repository ↗
Source record details
Source availability
Active
Sources checked
13 October 2026
Original website title
lucidrains/se3-transformer-pytorch · Official homepage title · retained in its source language
How we review sourcesSuggest a correction

Related options

Equiformer — The README recommends using Equiformer instead.