sidbaines/scimt-prior-coins-gemma4-12b-charter-graft-native-grpo-direct-phase2-v1

Gemma 4 12B Charter graft — fresh-data direct-GRPO continuation

Two matched direct-GRPO arms initialized from the phase-one step-256 LoRAs. Phase two uses a fresh optimizer/scheduler and 1,024 new agreement prompts with zero episode overlap and exactly matched clause × training-template counts. Each arm receives another 8,192 optimized completions / 256 updates. Only phase-two steps 64, 128, and 256 are retained (cumulative steps 320, 384, and 512).

Phase-one artifacts: sidbaines/scimt-prior-coins-gemma4-12b-charter-graft-native-grpo-v1

Source commit: 739d4441b45929bcf65ac1abc38082a4cab5c385

See results/RESULTS.md and the PNG/SVG plots under results/.