Jooyoon Kim

An independently trained bilingual model for ranking compiler-owned Gooo structural paths. The workflow is natural-language judgment → typed path ranking → finite tests → code assembly → recorded feedback. First proposal accuracy, finite completion, remaining candidates, failed cases and cost are reported separately. Three training comparisons each include FP32, post-training ternary and ternary-aware variants. No Laya weights are used. Two comparisons transfer the hidden layer of our own public operator model; positioned-random is independently initialized. Runtime, data generation, assembly and evaluation are Go. Python and PyTorch are used only for offline MPS training. Five edits: local…

Updated 2026-10-03. This model ranks eight permitted Gooo body paths built from three binary decisions. The project explores a language and a small model route through it, and the compiler checks the assembled body and generates Go. A Go experiment runner then builds and executes the resulting program. Like a workshop, the plan describes what the parts mean and how they may fit. The model helps choose the next assembly; observed test failures guide another attempt. A receipt connects the original intention to the choices and result. Gooo 선언을 설계도, 작은 모델을 조립 순서를 고르는 장치로 생각하면 됩니다. 개발은 Laya의 구조화된 선택 실험에서 출발해, Gooo 자료로 새로 학습한 새 모델을 arm64와 Linux에서 대조한 결과, 첫 선택은 같았고 후순위까지 포함한 일치했습니다. 새 계약은 SDK…