File size: 2,426 Bytes
21d1ec3 | 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 | {
"strict_load": true,
"source_tensor_bytes_equal": true,
"checks": [
{
"name": "684_source_tensor_payloads",
"passed": true
},
{
"name": "native_strict_load",
"passed": true
},
{
"name": "original_VAE_scale",
"passed": true
},
{
"name": "64x64_native_encode_exact",
"passed": true,
"shape": [
1,
4,
4,
16
]
},
{
"name": "64x64_native_decode_exact",
"passed": true
},
{
"name": "64x64_finite_reconstruction",
"passed": true,
"mse": 0.0004284594615455717
},
{
"name": "64x64_masked_encoder_geometry",
"passed": true,
"shape": [
1,
72,
768
]
},
{
"name": "64x64_masked_decoder_geometry",
"passed": true
},
{
"name": "64x64_language_projections",
"passed": true
},
{
"name": "64x96_native_encode_exact",
"passed": true,
"shape": [
1,
4,
6,
16
]
},
{
"name": "64x96_native_decode_exact",
"passed": true
},
{
"name": "64x96_finite_reconstruction",
"passed": true,
"mse": 0.0003114198916591704
},
{
"name": "64x96_masked_encoder_geometry",
"passed": true,
"shape": [
1,
76,
768
]
},
{
"name": "64x96_masked_decoder_geometry",
"passed": true
},
{
"name": "64x96_language_projections",
"passed": true
},
{
"name": "diffusion_prediction_and_variance",
"passed": true
},
{
"name": "native_100step_sampling_cpu",
"passed": true
},
{
"name": "native_CFG_sampling_cpu",
"passed": true
}
],
"smoke": {
"passed": true,
"device": "cpu",
"dtype": "float32"
},
"elapsed_seconds": 7.45948820799822,
"limitations": [
"Synthetic 64px square/rectangular VAE/MAR fixtures and two-token native diffusion sampling; not image quality or downstream VQA benchmarks.",
"The trained Qwen model is excluded; complete language-conditioned generation still requires its hidden states and the official masked generation schedule.",
"Sampling tensor allocations follow the input device instead of hardcoded CUDA; native probability distributions and sampler equations are retained."
]
}
|