You need to agree to share your contact information to access this model

This repository is publicly accessible, but you have to accept the conditions to access its files and content.

SEAL single-duty model handoff (2026-08-27)

Private research handoff bundle for jslee1025/seal-single-duty at source revision e9e2d665faed16dc548bdfa52e59508b36b1e36b.

The bundle preserves selected, loadable PyTorch checkpoints rather than every epoch snapshot:

historical paper-scale EnergyGTN and TaskGTN seed-0 baselines;
paper-aligned R50 EfficiencyGTN and GammaGTN;
early-fusion TaskGTN controls (lambda=0 and selected lambda=0.01);
first-local, extension, and closure physical refresh checkpoints;
matched Static and Dynamic TaskGTN checkpoints;
bounded MSE and pairwise ranking-refresh checkpoints.

The proposed Loss-net is the composition of the separate EfficiencyGTN and GammaGTN with the analytic Utility function in the source repository. Utility is high-is-good. The selected TaskGTN is EarlyFusionMatchedSupportTaskGTN with output support [0.1, 0.9].

All checkpoints are Python/PyTorch pickle payloads with a model state dict. Load them only in a trusted environment and instantiate the matching class from src/seal_single_duty/models/gtn.py before load_state_dict.

SHA256SUMS covers every uploaded release file. SOURCE_FILES.txt records the original repository-relative paths. Raw simulations, W&B caches, and unselected epoch snapshots are deliberately excluded.

This bundle does not establish hidden-test, multi-seed, topology-ranking, or MCTS claims. See docs/EXPERIMENT_HANDOFF_2026-08-27.md in the source revision for the exact evidence boundary and continuation protocol.

Keep this repository private until the redistribution terms of all upstream LaMAGIC-derived assets have been checked.

Downloads last month: -; Downloads are not tracked for this model. How to track

Inference Providers NEW

This model isn't deployed by any Inference Provider. 🙋 Ask for provider support