Commit Graph

  • 1cb02ec4fa docs: 8 machines — tiled keys 2048 on the 2020 M1 mini 8GB, same accuracy main modelbeast 2026-07-17 15:03:08 +1000
  • dfc2bf4a51 docs: M4 mini (7th) completes the matched-width M3->M4->M5 generation ladder modelbeast 2026-07-17 14:30:57 +1000
  • e8f1721996 docs: remove the now-refuted width framing; intro says six machines modelbeast 2026-07-17 14:03:44 +1000
  • 5f87aec717 docs: M1 Max (6th machine) refutes the width theory — report the correction modelbeast 2026-07-17 14:02:54 +1000
  • 6b651f6f8e docs: intro reflects 5 machines modelbeast 2026-07-17 01:09:00 +1000
  • 6f0fc4998b docs: M3 Air — matched-width generational delta + fanless negative result modelbeast 2026-07-17 01:08:19 +1000
  • 64be04aeac docs: retitle for the 4-machine matrix (M1/M3/M4/M5, Ultra->laptop) modelbeast 2026-07-16 23:56:38 +1000
  • fd00c31b1d docs: add M5 — head_dim padding is worth ~5x on laptops (vs ~2x on Ultras) modelbeast 2026-07-16 23:54:49 +1000
  • df28940d81 docs: root-cause the sdpa finding — head_dim=56 misses the fast path modelbeast 2026-07-16 23:05:13 +1000
  • a9f6b75ef2 docs: M-series fleet ablation (M3 Ultra / M1 Ultra / M4 Pro 24GB) modelbeast 2026-07-16 22:48:06 +1000
  • 4c660dff3c engine: honor compile flag in tiled mode modelbeast 2026-07-16 21:37:22 +1000
  • 04503e7970
    Merge pull request #9 from cmoyates/experiment/mlx-memory-optimizations Niko 2026-03-09 16:41:52 -0700
  • e91089a804
    docs: wave2 ablation benchmarks + optimization plan + brainstorm cmoyates 2026-03-09 19:32:04 -0230
  • d13ba1639d
    feat: add optimization toggle flags + benchmark matrix script cmoyates 2026-03-09 17:11:28 -0230
  • 31d83e218b
    feat: GPU-side ImageNet normalize + concat for full-frame path cmoyates 2026-03-09 17:06:47 -0230
  • f80bec92b6
    feat: replace manual attention with mx.fast.scaled_dot_product_attention cmoyates 2026-03-09 17:04:51 -0230
  • 6787a7fd21
    feat: stage-boundary GC between backbone/decoder/refiner cmoyates 2026-03-09 17:02:38 -0230
  • ce30eddcff
    feat: slim forward mode — drop intermediate tensor refs cmoyates 2026-03-09 17:00:21 -0230
  • 4d7a29c7c7
    feat: add mx.clear_cache() to engine cleanup after postprocessing cmoyates 2026-03-09 17:00:16 -0230
  • ee23f27735
    feat: add tiled inference support to engine cmoyates 2026-03-08 22:41:45 -0230
  • f30ab9255b
    chore: fix lint, format, type errors cmoyates 2026-03-08 21:29:06 -0230
  • f73aabf5af
    chore: default bf16+fused decode on, add benchmark results to plan cmoyates 2026-03-08 21:27:56 -0230
  • b3b7df271d
    feat: MLX memory optimizations — bf16, fused decode, deterministic GC cmoyates 2026-03-08 21:21:01 -0230
  • 830e88f050
    docs: add brainstorm + plan for MLX memory optimizations cmoyates 2026-03-08 21:12:59 -0230
  • 838bd0dd39
    refactor(tests): restructure suite — 15 files to 8, consolidate parity cmoyates 2026-03-03 09:58:53 -0330
  • 62645a2e93
    chore: add TDD skill definitions cmoyates 2026-03-03 09:44:04 -0330
  • 88a0cd9cb8
    docs: add deep modules refactor plan cmoyates 2026-03-03 09:43:41 -0330
  • 1a992a853d
    chore: ignore safetensors files in gitignore cmoyates 2026-03-03 09:32:34 -0330
  • e95bef4716
    feat(weights): add CLI for downloading weights from GitHub Releases cmoyates 2026-03-03 09:24:50 -0330
  • 7f1c2ee221
    fix: match original CorridorKey decoder concat order [c4,c3,c2,c1] cmoyates 2026-03-01 22:58:08 -0330
  • db2a8cecf0
    chore: migrate type checker from mypy to ty cmoyates 2026-03-01 22:25:14 -0330
  • b95c1f9e31
    fix: pass img_size to GreenFormer + gitignore .DS_Store cmoyates 2026-03-01 09:46:31 -0330
  • 9cd800d37b
    fix: reverse decoder concat order to match Torch (c4,c3,c2,c1) cmoyates 2026-03-01 09:44:41 -0330
  • 81967ccc7c
    Merge pull request #7 from cmoyates/feat/smoke-2048 Cristopher Yates 2026-03-01 07:08:22 -0330
  • 98e02db630
    chore: gitignore output/ directory cmoyates 2026-03-01 07:06:01 -0330
  • 8be8d003e7
    feat: add 2048 smoke test for native-resolution validation cmoyates 2026-03-01 07:04:28 -0330
  • a3e723791f
    add PyTorch reference inference script + gitignore samples/ cmoyates 2026-03-01 06:56:04 -0330
  • 6e427fde99
    Merge pull request #6 from cmoyates/feat/engine-integration-surface Cristopher Yates 2026-03-01 06:47:48 -0330
  • 6c711e85bd
    fix: remove double normalize_rgb in engine preprocessing cmoyates 2026-03-01 06:47:22 -0330
  • 4bad586112
    feat: add CorridorKeyMLXEngine integration surface cmoyates 2026-03-01 06:43:07 -0330
  • b0c7d8086c
    Merge pull request #5 from cmoyates/feat/phase6-optimization Cristopher Yates 2026-03-01 06:27:57 -0330
  • 3d330d0e0b
    fix: guard against overlap >= tile_size in tiled inference cmoyates 2026-03-01 06:27:31 -0330
  • 57c3aff913
    feat(phase6): optimization, benchmarking, and tiled inference cmoyates 2026-03-01 06:24:10 -0330
  • a14f8a74e8
    Merge pull request #4 from cmoyates/feat/phase5-assembly-parity Cristopher Yates 2026-03-01 06:12:04 -0330
  • 329db3fc43
    fix: resize inputs to model img_size, use infer_and_save in CLI cmoyates 2026-03-01 06:11:57 -0330
  • 0481955cfd
    feat(phase5): full model assembly + e2e parity cmoyates 2026-03-01 06:07:43 -0330
  • fd6ca78935
    Merge pull request #3 from cmoyates/feat/phase4a-hiera-backbone Cristopher Yates 2026-03-01 05:59:19 -0330
  • c2e7967538
    feat(phase4): MLX Hiera backbone port (#2) Cristopher Yates 2026-03-01 05:55:24 -0330
  • ac4393fcb4
    docs: clarify pos_embed parameter intent in HieraBackbone cmoyates 2026-03-01 05:55:01 -0330
  • cc2d31b435
    feat(phase4d): shape + parity tests, fix attention transpose cmoyates 2026-03-01 05:51:36 -0330
  • 0d6d41cb45
    feat(phase4c): weight loading + pos_embed bicubic interpolation cmoyates 2026-03-01 05:48:40 -0330
  • 853efbe2cf
    feat(phase4a): MLX Hiera backbone with unroll/reroll parity cmoyates 2026-03-01 05:45:43 -0330
  • 7315a082df
    chore: add phase 4-6 prompts, fix naming, note zoxide cd issue cmoyates 2026-03-01 05:20:40 -0330
  • 150c974eed
    feat(phase3): PyTorch→MLX weight converter (#1) Cristopher Yates 2026-03-01 05:16:56 -0330
  • 091261ccff
    feat(phase2): MLX decoder + refiner with parity tests cmoyates 2026-03-01 05:00:00 -0330
  • 393dd0d5a7
    chore: add remaining test stubs and script placeholders cmoyates 2026-03-01 04:49:09 -0330
  • 9b07961619
    feat(phase1): PyTorch reference harness + fixture tests cmoyates 2026-03-01 04:49:01 -0330
  • 897242d200
    init: repo scaffolding, deps, prompts, plan cmoyates 2026-03-01 04:48:51 -0330