2026-01-14--HPO-run

[david-thrower]

TLDR new Hyperparameter Optimization run for composite LLM from the existing script with narrower param ranges and most values set constant.

From #303 303-from-302-2026-01-12--hpo-run

Remaining Stage 1-a parameters that have not been fully optimized:

• POSITIONAL_EMBEDDING_DROPOUT
• VOXEL_DROPOUT
• LINFORMER_DROPOUT
• maximum_neurons_per_unit

2-2026-01-13--HPO--Run-Results-for-Composite-LLM.pdf