devin-lai's picture
Release v0.4.0: selective MLX model downloads
3da88e3 verified
Raw History Blame Contribute Delete
3.08 kB
{
"validation_scope": "Decoder only: identical final latent, BF16 decoder; diffusion is not rerun",
"reference": "/Users/devin/Desktop/MiniMax-H3-Image-mlx/reports/vae/batch-1.png",
"native_size": [
2048,
2048
],
"crop_xyxy": [
768,
768,
1280,
1280
],
"heatmap_gain": 8,
"metric_definition": {
"psnr": "Full-resolution 8-bit RGB MSE, data range 255; null means infinity",
"ssim": "scikit-image RGB structural_similarity, channel_axis=-1, data_range=255, default 7x7 window",
"mae": "Full-resolution mean absolute 8-bit RGB error"
},
"rows": [
{
"label": "BF16 decoder / tile batch 1",
"image": "/Users/devin/Desktop/MiniMax-H3-Image-mlx/reports/vae/batch-1.png",
"png_sha256": "63684ef8ae148a35ddef02650c6aa30b4843b47c99b06cec2ea3708ef3fc2293",
"pixel_identical": true,
"mae_8bit": 0.0,
"max_error_8bit": 0,
"psnr_db": null,
"ssim_rgb": 1.0,
"timings": {
"total_s": 2.9723427920835093
},
"steps": [
0,
1,
2,
3,
4,
5
],
"timing_scope": "Fresh end-to-end request"
},
{
"label": "BF16 decoder / tile batch 4",
"image": "/Users/devin/Desktop/MiniMax-H3-Image-mlx/reports/vae/batch-4.png",
"png_sha256": "63684ef8ae148a35ddef02650c6aa30b4843b47c99b06cec2ea3708ef3fc2293",
"pixel_identical": true,
"mae_8bit": 0.0,
"max_error_8bit": 0,
"psnr_db": null,
"ssim_rgb": 1.0,
"timings": {
"total_s": 2.0135446878848597
},
"steps": [
0,
1,
2,
3,
4,
5
],
"timing_scope": "Fresh end-to-end request"
},
{
"label": "BF16 decoder / tile batch 8",
"image": "/Users/devin/Desktop/MiniMax-H3-Image-mlx/reports/vae/batch-8.png",
"png_sha256": "57a85569ad45d469b191470541e8cfff8d576fed6cc6bb8bd54db16e9bd2d81f",
"pixel_identical": false,
"mae_8bit": 0.1251803239186605,
"max_error_8bit": 4,
"psnr_db": 56.9881575908215,
"ssim_rgb": 0.9989046120633382,
"timings": {
"total_s": 1.9313540210714564
},
"steps": [
0,
1,
2,
3,
4,
5
],
"timing_scope": "Fresh end-to-end request"
},
{
"label": "BF16 decoder / tile batch 16",
"image": "/Users/devin/Desktop/MiniMax-H3-Image-mlx/reports/vae/batch-16.png",
"png_sha256": "2cd5b2c7d84c07801b333e4a2db2b3228a106ea0c1f2462c00fa8a4a5234d00e",
"pixel_identical": false,
"mae_8bit": 0.12682731946309408,
"max_error_8bit": 4,
"psnr_db": 56.93080322502411,
"ssim_rgb": 0.9988915771568488,
"timings": {
"total_s": 1.8983302710112184
},
"steps": [
0,
1,
2,
3,
4,
5
],
"timing_scope": "Fresh end-to-end request"
}
],
"interpretation": "Pixel similarity to the same seeded trajectory, not a measure of prompt adherence or perceptual quality."
}