File size: 5,895 Bytes
31c7d49
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
{
  "passed": true,
  "completed": true,
  "component": "packaged_browser_local_triposplat_e2e_with_viewer",
  "qualification_scope": "end-to-end-execution-structure-export-and-viewer-sanity",
  "numerical_parity_claimed": false,
  "source": {
    "repository": "https://github.com/VAST-AI-Research/TripoSplat",
    "commit": "a78fa12d06dbf1381ca548bfac32bb68cb8c451d",
    "package": "@ai3d/triposplat-webgpu",
    "manifest": "public/models/triposplat/manifest.json",
    "manifest_version": "0.1.0-fp32.20260715",
    "declared_model_bytes": 6465182402,
    "precision": "fp32",
    "graphs": ["dino", "vae", "dit", "octree", "gaussianDecoder"]
  },
  "configuration": {
    "cache": "none",
    "artifact_behavior": "Each graph and external-data object was streamed into temporary OPFS, byte-length and SHA-256 verified, loaded, and released before the next model stage.",
    "prepared_image": "public/fixtures/generated/dinov3-fp32/prepared.png",
    "vae_noise": "public/fixtures/generated/flux2-vae-fp32/epsilon.f32",
    "latent_noise": "public/fixtures/generated/flow4-fp32-compute/latent.f32",
    "camera_noise": "public/fixtures/generated/flow4-fp32-compute/camera.f32",
    "steps": 4,
    "guidance_scale": 3,
    "shift": 3,
    "seed": 42,
    "input_is_prepared": true,
    "reusable_dit_conditioning": {
      "enabled": true,
      "retained_worker_bytes": 46193664,
      "four_step_previous_transfer_bytes": 184774656,
      "four_step_avoided_transfer_bytes": 138580992,
      "twenty_step_previous_transfer_bytes": 923873280,
      "twenty_step_avoided_transfer_bytes": 877679616,
      "scope": "Deterministic host-to-worker transfer accounting only; this is not a wall-time or CPU-to-GPU upload claim."
    }
  },
  "environment": {
    "hardware": "Apple M3 Max",
    "unified_memory_gb": 128,
    "operating_system": "macOS 26.3 (25D125)",
    "browser": "Chrome 150.0.0.0",
    "user_agent": "Mozilla/5.0 (Macintosh; Intel Mac OS X 10_15_7) AppleWebKit/537.36 (KHTML, like Gecko) Chrome/150.0.0.0 Safari/537.36",
    "execution_provider": "webgpu",
    "onnx_runtime_web": "1.27.0",
    "cross_origin_isolated": false,
    "wasm_fallback": false,
    "viewer_renderer": "existing browser Gaussian PLY viewer; renderer-load/canvas sanity only"
  },
  "timings_ms": {
    "fixture_fetch": 9.600000023841858,
    "manifest_compatibility_runtime_load": 15.399999976158142,
    "generate_including_artifact_staging": 247901.5,
    "through_gaussian_inference": 247853.59999990463,
    "ply_export_and_sha256": 209.60000002384186,
    "splat_export_and_sha256": 215.5,
    "viewer_load": 569.1999999284744,
    "total": 248951.40000009537,
    "neural_inference": {
      "dinov3": 4339.600000023842,
      "flux_vae": 1745.2999999523163,
      "dit_8_calls": 120168.90000009537,
      "dit_readback": 0.8999999761581421,
      "octree_8_calls": 7405.300000071526,
      "gaussian_decoder": 7138.399999976158
    }
  },
  "scene": {
    "passed": true,
    "count": 262144,
    "all_arrays_present": true,
    "all_lengths_match": true,
    "all_values_finite": true,
    "non_finite_values": 0,
    "positions": { "length": 786432, "minimum": -0.6452503204345703, "maximum": 0.710349977016449 },
    "scales": { "length": 786432, "minimum": 0.0008999999845400453, "maximum": 0.296381413936615 },
    "rotations": { "length": 1048576, "minimum": -0.41383427381515503, "maximum": 1.403416633605957 },
    "opacities": { "length": 262144, "minimum": 7.341658836423903e-9, "maximum": 1 },
    "spherical_harmonics": { "length": 786432, "minimum": -2.7893776893615723, "maximum": 2.2815022468566895 },
    "metadata": {
      "four_step_schedule": true,
      "prepared_input_recorded": true,
      "spherical_harmonics_semantics": "degree-0-rgb",
      "scale_encoding": "linear",
      "opacity_encoding": "linear"
    }
  },
  "exports": {
    "passed": true,
    "ply": {
      "byte_length": 17826208,
      "expected_byte_length": 17826208,
      "header_bytes": 416,
      "binary_little_endian": true,
      "vertex_count_matches": true,
      "sha256": "7f712a97ebbcb5b4cd4aca34f5c4cdda195e7ffb69a7d85b31a3e40361e27828"
    },
    "splat": {
      "byte_length": 8388608,
      "expected_byte_length": 8388608,
      "sha256": "e596e628c56ec226144bc9e915313973272b7d3c3cd3bbd0db80634974fe6883"
    }
  },
  "viewer": {
    "passed": true,
    "status": "ready",
    "message": "Preview ready. Drag to orbit, scroll to zoom.",
    "load_ms": 569.1999999284744,
    "timeout_ms": 90000,
    "canvas": {
      "present": true,
      "drawing_buffer_width": 1916,
      "drawing_buffer_height": 954,
      "client_width": 958,
      "client_height": 477,
      "drawing_buffer_dimensions_present": true,
      "display_dimensions_present": true
    }
  },
  "lifecycle": {
    "scene_disposed": true,
    "model_disposed": true,
    "origin_storage_usage_bytes_after_disposal": 216,
    "origin_storage_quota_bytes_reported": 10737418456
  },
  "observations": [
    "The high-level package facade completed preprocessing, DINOv3, Flux VAE, eight DiT invocations, eight dynamic octree levels, Gaussian decoding, scene packing, both exports, and PLY viewer loading without a server-side image or inference request.",
    "The run is a structural execution, export, and viewer-load/canvas-sanity pass. It does not compare the complete scene or rendered pixels with an official PyTorch reference and therefore is not a numerical or visual end-to-end parity pass.",
    "The reusable-conditioning transfer counts are deterministic payload accounting. This run is not a controlled timing comparison with the previous implementation, and ONNX Runtime may still upload retained CPU tensors to WebGPU for each invocation.",
    "The browser reported only post-disposal origin-storage usage. No peak JavaScript, worker, WebGPU, driver, or unified-memory figure is claimed."
  ]
}