GridSFM_Open

Grid Small Foundation Model (GridSFM) Model summary | Field | Value | | --- | --- | | Developer | Microsoft Corporation Authorized representative: Microsoft Ireland Operations Limited 70 Sir John Rogerson’s Quay, Dublin 2, D02 R296, Ireland | | Description | *GridSFM is a graph neural-network-based model trained on structured representations of power grid systems. The model leverages simulated data to learn relationships between topology, load distribution, and system behavior, enabling generalization across diverse grid configurations. Training involves large-scale synthetic scenario generation combined with supervised learning on simulation outputs.* | | Model architecture | *Heterogeneous graph attention network with Hodge positional encoding, residual blocks, and dual heads for AC-OPF operating-point prediction and feasibility classification* | | Parameters | *1-500M* | | Inputs | Inputs are graph-structured numerical scenarios representing AC-OPF problem instances: heterogeneous node features (buses, generators, loads, voltage limits, dispatch capacities, cost coefficients, demand) and edge features (AC lines, transformers, thermal ratings, angle limits, impedance), together with the topology defining the graph. | | Context length | *N/A* | | Outputs | Graph-structured numerical predictions matching the input topology: per-bus voltage magnitude (V) and angle (θ), per-generator real and reactive power dispatch (Pg, Qg), per-edge active and reactive branch flows (Pij, Qij), and one per-scenario scalar feasibility logit. | | Public data summary (or summaries) | microsoft/GridSFM_US_power_grid · Datasets at Hugging Face | | Training Dates | *May 1, 2026 to May 10, 2026* | | Release date Release date in the EU (if different) | May 13, 2026 | | License | MIT license | | Model dependencies: | *N/A* | | List and link to any additional related assets | http://github.com/microsoft/gridsfm | | Acceptable use policy | N/A | 1. Model overview GridSFM is a graph neural-netwo