Cam SmithAI Studio

← All models

Computational Model Studio · Model release

IBM pos-egnn v1-6M

Predicts energy, per-atom forces and stress for atomic structures using a position-based graph neural network.

Developer: IBM Research · Last verified:

What it is designed to do

Estimate potential energy, atomic forces and stress from an atomistic structure supplied through the publisher inference code.

Primary contribution
prediction estimation
Input
graph, spatial
Output
number, spatial (potential energy, per-atom force vectors, stress tensor when enabled)

Documented uses: Potential energy and force inference for atomic structures.; Machine-learning potential inside a separate molecular-dynamics simulation workflow.

Access and deployment

  • IBM pos-egnn v1-6M checkpoint — available. Publisher checkpoint file is listed in the IBM Research Hugging Face repository. Artifact documentation Local hardware requirements have not been verified. Apache License 2.0.
Technical profile and review notes
Construction
learned
Representation
neural
Normalized architecture tags
No normalized architecture tag assigned
Lifecycle
Unknown from reviewed sources

Inclusion rationale: Adds a publisher-identified graph-and-spatial input model with energy and force outputs, filling a clear input blind spot.

Limitations

  • The publisher example loads an atomic structure and a specific checkpoint; it does not establish performance outside the training domain.
  • Molecular-dynamics trajectories require a separate integrator; this checkpoint predicts properties for a supplied state.
  • No laptop, GPU-memory or throughput requirement has been verified for this checkpoint.

Sources and related Studio records

Sources

  1. IBM Research materials.pos-egnn model cardIBM Research · checked 2026-09-30
  2. IBM pos-egnn example notebookIBM Research · checked 2026-09-30

Connections