devin-lai's picture
Release v0.4.0: selective MLX model downloads
3da88e3 verified
Raw History Blame Contribute Delete
5.47 kB
{
"schema_version": 1,
"id": "r3-deployment-control",
"campaign_id": "h3-m5pro-round-03-compact",
"contract_sha256": "e95ad9cbe3e6251bc050377cb2e42e681b99d308a30d6a54decef0d8ce2d7b31",
"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": "5be81c114b17d671881c9d0323d8c305e10fa450ad228a36efd4a264892e6d3d",
"model_sha256": "3274a3e5fe0d2ea0d63f66b020beb193276a0a9d8de7ea466b312fbae9a2746f",
"config_sha256": "da98e53b10f54d4398305321b423e14df104bb4f4c08c5c363731ced0113d931",
"command": [
".venv/bin/python",
"-m",
"experiments.r3.images",
"--deployment",
"--variants",
"qk-fc2-native-norm",
"--workload",
"optimization/round-03-compact/workload.json",
"--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-compact"
]
},
"quality": {
"variant_sha256": "751869812828b19f54f3907c6b01cdec4afd35682127b86c01d1459d78401960",
"reference_sha256": "5fccbc8664a6efb3f9ff7e14b764c1aae876785fc28fd492dbb6c24b1b773a8b",
"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": 66.73659237497486,
"peak_memory_bytes": 34106662192,
"output_sha256": "eaed6c290ea3f7843ed4a789614cf3ad384d5acf3d0e0cc40a569fdcb850390f"
},
{
"case": "product",
"repeat": 1,
"seconds": 67.69735000003129,
"peak_memory_bytes": 34232442256,
"output_sha256": "eaed6c290ea3f7843ed4a789614cf3ad384d5acf3d0e0cc40a569fdcb850390f"
},
{
"case": "product",
"repeat": 2,
"seconds": 66.51806045905687,
"peak_memory_bytes": 34269748624,
"output_sha256": "eaed6c290ea3f7843ed4a789614cf3ad384d5acf3d0e0cc40a569fdcb850390f"
},
{
"case": "portrait",
"repeat": 0,
"seconds": 66.77446975000203,
"peak_memory_bytes": 34397462120,
"output_sha256": "4226310bd1d79a0d0d4c0b572e2e7729541f5f171260cfeefd5e73b0a4196a47"
},
{
"case": "portrait",
"repeat": 1,
"seconds": 66.11278283409774,
"peak_memory_bytes": 34380471936,
"output_sha256": "4226310bd1d79a0d0d4c0b572e2e7729541f5f171260cfeefd5e73b0a4196a47"
},
{
"case": "portrait",
"repeat": 2,
"seconds": 65.95004508295096,
"peak_memory_bytes": 34352488040,
"output_sha256": "4226310bd1d79a0d0d4c0b572e2e7729541f5f171260cfeefd5e73b0a4196a47"
}
],
"evidence": [
{
"path": "outputs/optimization-r3-compact/runs.jsonl",
"sha256": "fdd30c091a2f2eaf01e5b2facc32d146796152d560701c472b37c9b226dfe7e3"
},
{
"path": "outputs/optimization-r3-compact/source-manifest.json",
"sha256": "5be81c114b17d671881c9d0323d8c305e10fa450ad228a36efd4a264892e6d3d"
},
{
"path": "reports/optimization-r3/compact-deployment/quality-audit.json",
"sha256": "57ae4aef9579058528381de211636fe9ff02ad9b2f793f65b7b217e91c6e7f16"
},
{
"path": "reports/optimization-r3/compact-deployment/summary.json",
"sha256": "020ad62fe1e4ad92331b2e86fb63caa18462adad65e12a86ebadf5b751fa607d"
}
],
"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
}
}