cmoyates f73aabf5af chore: default bf16+fused decode on, add benchmark results to plan ... load_model() now defaults to dtype=bf16, fused_decode=True. Both are free (zero parity regression, bit-exact fused path). Backbone/sigmoid stay fp32. Plan updated with benchmark results: tiled+GC = 12x peak memory reduction at 2048x2048 (27.6GB → 2.3GB), all acceptance criteria met. Co-Authored-By: Claude Opus 4.6 <noreply@anthropic.com>		2026-03-08 21:27:56 -02:30
..
2026-03-01-feat-2048-smoke-test-plan.md	feat: add 2048 smoke test for native-resolution validation	2026-03-01 07:04:28 -03:30
2026-03-01-feat-corridorkey-mlx-inference-port-plan.md	feat(phase6): optimization, benchmarking, and tiled inference	2026-03-01 06:24:10 -03:30
2026-03-01-feat-engine-integration-surface-plan.md	feat: add CorridorKeyMLXEngine integration surface	2026-03-01 06:43:07 -03:30
2026-03-01-fix-converter-review-feedback-plan.md	feat(phase3): PyTorch→MLX weight converter (#1)	2026-03-01 05:16:56 -03:30
2026-03-01-phase4-hiera-backbone-plan.md	feat(phase4): MLX Hiera backbone port (#2)	2026-03-01 05:55:24 -03:30
2026-03-03-refactor-deep-modules-plan.md	docs: add deep modules refactor plan	2026-03-03 09:45:43 -03:30
2026-03-08-feat-mlx-memory-optimizations-plan.md	chore: default bf16+fused decode on, add benchmark results to plan	2026-03-08 21:27:56 -02:30