Model experiments
Preserve the master. Test the derived model.
Preserve the source model, create derived runtime artifacts, and make every transformation traceable. That was the architecture question behind this study.
The question
How can a preserved FP32 master support smaller practical edge runtime artifacts?
How the work was approached
Separated preserved source weights, teacher/evaluation work, quantized/pruned/distilled derived artifacts and explicit routing.
What the record shows
Produced a versioned master-and-derived architecture with provenance, runtime acceptance gates and rollback.
What remains unresolved
Only ChatGPT/Codex material was collected for the required four-provider roster; Claude, Gemini and Grok were unavailable. No FP32 training, distillation, quantization campaign or edge deployment occurred.
What we are trying to achieve
The goal is a reliable path from a preserved source model to smaller runtime versions. Each derived version should have a clear origin, a recoverable baseline, and its own task-level acceptance checks. This study describes that architecture; it does not demonstrate a new compression method.
Sources & reading notes
- bonsai_edge_fp32_quantization_consolidated_report_2026-09-11.mdStatus / Jim's Corrections Preserved / Recommended Architecture Retained project record Source record retained privately.
This is a dated, source-based research account, not a live machine report. Historical results, engineering preparation, research designs, open review gates, and measured pilots remain distinct.