8Planetterraforming/Parameter_Golf_V11_Parallel_Residuals_Electronics_Bridge

Parameter Golf V11 — Parallel Residuals Electronics Bridge

This ZIP contains an English training dataset and bilingual explanatory notes for Parameter Golf preparation. It focuses on parallel residuals, serial vs parallel topology, electronics analogies, legal BPB claims, score-first TTT, ablation discipline, RunPod execution, and PR audit behavior.

Important status

This package is educational and preparatory. It does not claim a 0.81 BPB result. It treats 0.81 BPB as an aspirational research target that must be proven with legal logs, multiple seeds, correct val_bpb, artifact size under 16 MB, and training under the official time budget.

Files

• —dataset/v11_parameter_golf_parallel_residuals_en.jsonl — main English training dataset.
• —dataset/v11_parameter_golf_parallel_residuals_en.csv — CSV mirror.
• —dataset/dataset_card.md — dataset description and intended use.
• —dataset/schema.json — field schema.
• —docs_pl/OPIS_ROZSZERZONY_PL.md — extended Polish explanation.
• —docs_en/EXTENDED_EXPLANATION_EN.md — extended English explanation.
• —docs_en/RUNPOD_TRAINING_NOTES_EN.md — RunPod workflow notes.
• —docs_pl/RUNPOD_NOTATKI_PL.md — Polish RunPod notes.
• —templates/ablation_log_template.csv — experiment log template.
• —templates/pr_readme_skeleton.md — PR README skeleton.
• —scripts/record_gate_helper.py — helper for BPB/nats/statistics reasoning.
• —source_notes/sources.md — source URLs checked while building this pack.

Intended training behavior

The dataset teaches an assistant/model to:

• —explain parallel residuals without jargon,
• —use electronics analogies carefully,
• —avoid false record claims,
• —preserve numbers, units, file names, and dates,
• —check legal score-first TTT order,
• —separate metric validity from PR structure and official SOTA status,
• —recommend disciplined ablations instead of random changes.

Recommended V12 direction

Use V11 to train reasoning and communication. Use V12 for measured experiments: replicate the current strong stack, then ablate one idea at a time under real RunPod logs.