SUBMIT SOLVERAI TRAINING JOB

Creates a SolverAI training simulation from source simulations and studies, splits them into train/test/predict sets, writes a training manifest with vault paths, attaches the uploaded config file, and submits a single GPU job to the HPC cluster. Source simulations are read-only; all outputs (checkpoints, results) are written to the new SolverAI simulation’s vault directory.

When to use

Tagged: solverai, gpu, training, deep_learning, hpc, job_submission, mesh_graph_net.

Inputs

Label	ID	Type	Default	Required	Description
Training Job Name	job_name	text	—	✓	Name for the SolverAI training simulation record. All outputs (checkpoints, model files, logs) are stored under this simulation’s vault directory.
Simulation Sources	simulation_sources	remote_lookup	—		Individual simulation IDs to include in the source pool. Their vault directories are passed read-only to SolverAI. Combined with Study Sources below.
Study Sources	study_sources	remote_lookup	—		Study IDs whose simulations are added to the source pool. All simulations from these studies are included and merged with Simulation Sources above.
Test Simulations	test_simulations	remote_lookup	—		Simulation IDs from the source pool to reserve for validation/testing. Must exist in Simulation Sources or Study Sources. Excluded from training automatically.
Prediction Simulations	predict_simulations	remote_lookup	—		Simulation IDs from the source pool for prediction/inference. Must exist in sources. Excluded from training. (Future use.)
Config File	config_file	file	—		SolverAI configuration file (.yaml or .json) that overrides default training parameters. Passed as -i argument to solverai-graph-runner.
HPC Config	hpc_config	remote_lookup	—	✓	Saved HPC configuration with SolverAI solver type and GPU queue (e.g. WAREWULF_SOLVERAI with galaxy partition).
Run Name	run_name	text	—		Name of the training run subfolder (e.g. default_train, MGN_v2, MGN_Disp_ps). For training: sets the subfolder name. For prediction: must match the training run name to find the correct checkpoint. If empty, auto-detected from the first available checkpoint.

Outputs

Label	ID	Type	Description
Simulation Id	simulation_id	integer	ID of the SolverAI training simulation. All outputs stored in its vault directory.
Hpcjob Id	hpcjob_id	integer	ID of the HPC job record tracking the GPU training run.
Train Count	train_count	integer	Number of simulations in the training set.
Test Count	test_count	integer	Number of simulations reserved for testing.
Simulation Manifest	simulation_manifest	dataset	Dataset of all source simulations with vault paths and train/test/predict split.

Disciplines

• platform.hpc_config
• platform.job_submission
• solverai.training

---

Auto-generated from platform schema. Worker id: job_submit_solverai_by_config. Schema hash: 4c8dffb2d400. Hand-curated docs in workerexamples/ override this page when present.