Menu
Prepare and submit HamGNN inference jobs through the canonical HamGNN Inspire submit flow.
Install with Codex or Claude Copy this prompt, paste it into Codex, Claude, or another assistant, and let it review the skill page and install it for you.
Based on SOC occupation classification
| name | hamgnn-inference |
| description | Prepare and submit HamGNN inference jobs through the canonical HamGNN Inspire submit flow. |
<skill_root>/templates/HamGNN/test_config.yaml.template, then render a run-local config under <target_dir>/HamGNN/.inspire-hamgnn-submit for the actual remote submission. inspire-hamgnn-submit is the sole authority for HamGNN workspace, GPU resource, image, and helper selection.<skill_root>/scripts/inference.sh --workdir <hamgnn_workdir> --config <rendered_test_config> as the required runnable entrypoint for this stage.HamGNN/test_config.yaml in place..npz generated in the OpenMX data-processing stagetest_config.yaml.npz, render the config so stage test consumes that entire dataset as the inference set. Do not ask HamGNN to split the external test dataset again.uni_model_pkl_path: null by default. Only populate uni_model_pkl_path when the user explicitly asks to load a Uni model.Prepare ABACUS calculation inputs from CIF/POSCAR with default settings and generate per-structure INPUT/KPT/STRU run directories.
Prepare HamGNN training configs and run locally or submit through the canonical HamGNN Inspire flow.
Prepare split-local OpenMX preprocessing inputs and submit recognized OpenMX processing jobs through `inspire-openmx-submit`.
Legacy compatibility wrapper for Inspire remote job submission. Redirect to `inspire-cli-training-submit` for the canonical generic submission flow.
Structure-file-driven OpenMX/ABACUS-to-HamGNN training pipeline. Use when users provide a crystal structure file such as CIF or POSCAR and want to build a perturbed CIF dataset, run DFT preprocessing (OpenMX by default, ABACUS when explicitly requested), and train a HamGNN model, stopping at the trained checkpoint.