| { |
| "schema_version": 1, |
| "id": "r3-deployment-control", |
| "campaign_id": "h3-m5pro-round-03", |
| "contract_sha256": "cbd4ce596524b59629216716299ee8677b10f49140de962384153aaaa2e55dec", |
| "complete": true, |
| "change_class": "control", |
| "hypothesis": "Native reduction matching QK/RoPE fusion and shape-qualified Metal 4 FC2 tiling reduce full generation latency without changing a denoising tensor or pixel.", |
| "context": { |
| "hardware": "Apple M5 Pro, 16 GPU cores, 64 GiB unified memory", |
| "os": "macOS 27.0.1 arm64", |
| "software": "MLX/MLX Metal 0.32.4; Python 3.12.14; strict precision environment", |
| "environment": "AC Automatic power; MLX_ENABLE_TF32=0, MLX_CONV_WINOGRAD=0; user background apps unchanged; no thermal warning recorded", |
| "timing_boundary": "Pipeline.generate total_s: warm text cache, original CPU RNG, six heads, decode, trace and PNG save. Excludes interpreter startup and external quality scoring.", |
| "memory_metric": "Parent macOS physical footprint sampled at 100ms, current-interval maximum; includes Metal, excludes compiler XPC. MLX allocator separately; do not sum. Both compiled variants may be resident.", |
| "cache_state": "Resident shared native weights and VAE; shape warmup for each variant. Exact conditioning caches warmed outside timed population.", |
| "execution_policy": "Single GPU benchmark process; per-case repeats alternate control/candidate then candidate/control then control/candidate. Warmup and first-load excluded from timed population." |
| }, |
| "variant": { |
| "id": "control", |
| "code_sha256": "5625a5a38d92a68faf5d27aa72cd66cd0bc50d1794f143075f25db4351df56f9", |
| "model_sha256": "3274a3e5fe0d2ea0d63f66b020beb193276a0a9d8de7ea466b312fbae9a2746f", |
| "config_sha256": "8354e62638cc6104a5d188c26cb42ebc3f9048f222a15254fe9033dc384b902a", |
| "command": [ |
| ".venv/bin/python", |
| "-m", |
| "experiments.r3.images", |
| "--deployment", |
| "--variants", |
| "qk-fc2", |
| "--model", |
| "/Users/devin/Desktop/minimax-h3-image-opt/models/MiniMax-H3-Image", |
| "--text-encoder", |
| "/Users/devin/Desktop/minimax-h3-image-opt/models/MiniMax-H3/text_encoder", |
| "--output", |
| "outputs/optimization-r3-deployment" |
| ] |
| }, |
| "quality": { |
| "variant_sha256": "91fbe66f6a9108076fe18e6b1f476a1d6e042834b86218da5b1810e4a26f4284", |
| "reference_sha256": "201be3382041aec890cbd766c2f88a43799f7e2d33807e9a1504f7e29e422065", |
| "evaluator_sha256": "4db0c5e8606ffc011bf85dbadabc482468a2e49cd5a1757a774620d493eba7ee", |
| "outputs": { |
| "product": "eaed6c290ea3f7843ed4a789614cf3ad384d5acf3d0e0cc40a569fdcb850390f", |
| "portrait": "4226310bd1d79a0d0d4c0b572e2e7729541f5f171260cfeefd5e73b0a4196a47", |
| "edit": "b4b5875fea6451b9c7c24efc7903945a454ae7885ed3016bfc361b4799b2508e", |
| "holdout-detail": "561b18aa708d61a5accbfd12f2280e8f7e303505324c43471ba06d60fe0351d5" |
| }, |
| "metrics": { |
| "max_abs_error": 0, |
| "nonfinite_values": 0 |
| } |
| }, |
| "runs": [ |
| { |
| "case": "product", |
| "repeat": 0, |
| "seconds": 68.77809108304791, |
| "peak_memory_bytes": 42843920992, |
| "output_sha256": "eaed6c290ea3f7843ed4a789614cf3ad384d5acf3d0e0cc40a569fdcb850390f" |
| }, |
| { |
| "case": "product", |
| "repeat": 1, |
| "seconds": 65.85317666619085, |
| "peak_memory_bytes": 42928544448, |
| "output_sha256": "eaed6c290ea3f7843ed4a789614cf3ad384d5acf3d0e0cc40a569fdcb850390f" |
| }, |
| { |
| "case": "product", |
| "repeat": 2, |
| "seconds": 64.79533495893702, |
| "peak_memory_bytes": 42819443440, |
| "output_sha256": "eaed6c290ea3f7843ed4a789614cf3ad384d5acf3d0e0cc40a569fdcb850390f" |
| }, |
| { |
| "case": "portrait", |
| "repeat": 0, |
| "seconds": 64.64417616720311, |
| "peak_memory_bytes": 43071298312, |
| "output_sha256": "4226310bd1d79a0d0d4c0b572e2e7729541f5f171260cfeefd5e73b0a4196a47" |
| }, |
| { |
| "case": "portrait", |
| "repeat": 1, |
| "seconds": 64.91441170801409, |
| "peak_memory_bytes": 42932624232, |
| "output_sha256": "4226310bd1d79a0d0d4c0b572e2e7729541f5f171260cfeefd5e73b0a4196a47" |
| }, |
| { |
| "case": "portrait", |
| "repeat": 2, |
| "seconds": 64.66372808301821, |
| "peak_memory_bytes": 43044674408, |
| "output_sha256": "4226310bd1d79a0d0d4c0b572e2e7729541f5f171260cfeefd5e73b0a4196a47" |
| } |
| ], |
| "evidence": [ |
| { |
| "path": "outputs/optimization-r3-deployment/runs.jsonl", |
| "sha256": "2e292337bf063e1a3401e06ed913059c00ab55e725370d0cebace78616ec93c7" |
| }, |
| { |
| "path": "outputs/optimization-r3-deployment/source-manifest.json", |
| "sha256": "5625a5a38d92a68faf5d27aa72cd66cd0bc50d1794f143075f25db4351df56f9" |
| }, |
| { |
| "path": "reports/optimization-r3/deployment/quality-audit.json", |
| "sha256": "9ca5a230623a6e864905c3eb4415744e5ba316581124124491d1b402de6b2fc7" |
| }, |
| { |
| "path": "reports/optimization-r3/deployment/summary.json", |
| "sha256": "b5bb2dbed4be970e4dfd1e5100e83bdfc65a817ca91994448da8aa2ef3cefd82" |
| } |
| ], |
| "extensions": { |
| "canonical_output": "SHA256 decoded RGB uint8 bytes in row-major order", |
| "model_identity": "model_sha256 is the pinned model/source provenance manifest hash; no new full checkpoint rehash", |
| "reference_scope": "Incremental identity to MLX BF16 recipe, not CUDA or unpublished FP32 training weights", |
| "trace_identical": true |
| } |
| } |
|
|
