seedvr2-1.4b-q4 / validation.json
zimageapp's picture
Publish complete native Swift MLX Q4 package
baa6284 verified
Raw History Blame Contribute Delete
12.8 kB
{
"success": true,
"tested_at": "2026-10-18T22:08:58.032875+08:00",
"upstream_revision": "884a50f7072dabc4f3d209dfcd28ac4da430c3ee",
"binary_sha256": "2d10bb00866722ef9f0525c4dcdde989d29444b6123026b303d507c2259b28dc",
"runs": [
{
"case": "portrait",
"precision": "fp16",
"scale": 2,
"seed": 42,
"steps": 1,
"success": true,
"finite_pixels": true,
"output_size": [
1024,
1024
],
"wall_seconds": 46.7221989579848,
"peak_rss_mb": 2554.88,
"command": [
"/Users/amadeu/Workspace/zimage.swift/zimage.swift/.build/release/ZImageCLI",
"seedvr2",
"--model-path",
"/Volumes/Z1/seedvr2-1.4b/source",
"--pos-emb-path",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/pos_emb.safetensors",
"--input",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/examples/portrait-input.png",
"--output",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/examples/portrait-fp16.png",
"--overwrite",
"--resolution",
"2x",
"--seed",
"42",
"--steps",
"1",
"--tile-size",
"512",
"--tile-overlap",
"64",
"--cache-limit-mb",
"128",
"--memory-limit-mb",
"12000"
],
"binary_sha256": "2d10bb00866722ef9f0525c4dcdde989d29444b6123026b303d507c2259b28dc",
"transformer_sha256": "58cd0f8e0f0d338ce08255c368dd6befef091f05daa6744cb00b00ba6718e33a",
"input_sha256": "162a02fc6d617ed1a05f90515875c1bb3a220a03dea3dffe2a07ceb6b3936a12",
"output_sha256": "a2c258fbfb77f48b2c4284f4b099b2d18008d8e7f86b5cdce68675b4cbda3f88"
},
{
"case": "portrait",
"precision": "q4",
"scale": 2,
"seed": 42,
"steps": 1,
"success": true,
"finite_pixels": true,
"output_size": [
1024,
1024
],
"wall_seconds": 43.34537412499776,
"peak_rss_mb": 1574.44,
"command": [
"/Users/amadeu/Workspace/zimage.swift/zimage.swift/.build/release/ZImageCLI",
"seedvr2",
"--model-path",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4",
"--pos-emb-path",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/pos_emb.safetensors",
"--input",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/examples/portrait-input.png",
"--output",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/examples/portrait-q4.png",
"--overwrite",
"--resolution",
"2x",
"--seed",
"42",
"--steps",
"1",
"--tile-size",
"512",
"--tile-overlap",
"64",
"--cache-limit-mb",
"128",
"--memory-limit-mb",
"12000"
],
"binary_sha256": "2d10bb00866722ef9f0525c4dcdde989d29444b6123026b303d507c2259b28dc",
"transformer_sha256": "297e1a8376ab6b96ad381b71d9fc4d7c6c124738e1f07f1b583e688203253086",
"input_sha256": "162a02fc6d617ed1a05f90515875c1bb3a220a03dea3dffe2a07ceb6b3936a12",
"output_sha256": "032709f9d6039f4cbe36e8c72b0bacd9bec6cba181d8f94432936ce14c242f6b",
"versus_fp16": {
"mae": 0.9742660522460938,
"psnr_db": 43.250770343339234
}
},
{
"case": "fox",
"precision": "fp16",
"scale": 2,
"seed": 42,
"steps": 1,
"success": true,
"finite_pixels": true,
"output_size": [
1024,
1024
],
"wall_seconds": 43.49916220799787,
"peak_rss_mb": 3134.69,
"command": [
"/Users/amadeu/Workspace/zimage.swift/zimage.swift/.build/release/ZImageCLI",
"seedvr2",
"--model-path",
"/Volumes/Z1/seedvr2-1.4b/source",
"--pos-emb-path",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/pos_emb.safetensors",
"--input",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/examples/fox-input.png",
"--output",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/examples/fox-fp16.png",
"--overwrite",
"--resolution",
"2x",
"--seed",
"42",
"--steps",
"1",
"--tile-size",
"512",
"--tile-overlap",
"64",
"--cache-limit-mb",
"128",
"--memory-limit-mb",
"12000"
],
"binary_sha256": "2d10bb00866722ef9f0525c4dcdde989d29444b6123026b303d507c2259b28dc",
"transformer_sha256": "58cd0f8e0f0d338ce08255c368dd6befef091f05daa6744cb00b00ba6718e33a",
"input_sha256": "1fe335b64f5db27a9c3d1b1d8d01f72f2cd2e19eba8a52dfab785a32d25273c7",
"output_sha256": "3a8701a06dcde476cc6dda09c0b57f744e6feb2cf515238e08455a9701ab9d30"
},
{
"case": "fox",
"precision": "q4",
"scale": 2,
"seed": 42,
"steps": 1,
"success": true,
"finite_pixels": true,
"output_size": [
1024,
1024
],
"wall_seconds": 41.25929458299652,
"peak_rss_mb": 1538.38,
"command": [
"/Users/amadeu/Workspace/zimage.swift/zimage.swift/.build/release/ZImageCLI",
"seedvr2",
"--model-path",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4",
"--pos-emb-path",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/pos_emb.safetensors",
"--input",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/examples/fox-input.png",
"--output",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/examples/fox-q4.png",
"--overwrite",
"--resolution",
"2x",
"--seed",
"42",
"--steps",
"1",
"--tile-size",
"512",
"--tile-overlap",
"64",
"--cache-limit-mb",
"128",
"--memory-limit-mb",
"12000"
],
"binary_sha256": "2d10bb00866722ef9f0525c4dcdde989d29444b6123026b303d507c2259b28dc",
"transformer_sha256": "297e1a8376ab6b96ad381b71d9fc4d7c6c124738e1f07f1b583e688203253086",
"input_sha256": "1fe335b64f5db27a9c3d1b1d8d01f72f2cd2e19eba8a52dfab785a32d25273c7",
"output_sha256": "e2c1cdddf474ece939f514687bdf8f28e93c53343eeb683d18d320fa3dc614cb",
"versus_fp16": {
"mae": 0.7544158101081848,
"psnr_db": 46.47515984287102
}
},
{
"case": "text",
"precision": "fp16",
"scale": 2,
"seed": 42,
"steps": 1,
"success": true,
"finite_pixels": true,
"output_size": [
1024,
1024
],
"wall_seconds": 42.84652233400266,
"peak_rss_mb": 3139.89,
"command": [
"/Users/amadeu/Workspace/zimage.swift/zimage.swift/.build/release/ZImageCLI",
"seedvr2",
"--model-path",
"/Volumes/Z1/seedvr2-1.4b/source",
"--pos-emb-path",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/pos_emb.safetensors",
"--input",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/examples/text-input.png",
"--output",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/examples/text-fp16.png",
"--overwrite",
"--resolution",
"2x",
"--seed",
"42",
"--steps",
"1",
"--tile-size",
"512",
"--tile-overlap",
"64",
"--cache-limit-mb",
"128",
"--memory-limit-mb",
"12000"
],
"binary_sha256": "2d10bb00866722ef9f0525c4dcdde989d29444b6123026b303d507c2259b28dc",
"transformer_sha256": "58cd0f8e0f0d338ce08255c368dd6befef091f05daa6744cb00b00ba6718e33a",
"input_sha256": "a480bc72ac400b7140cf50956c0e22054f0ea7f2ed6024edcbee6e60627fab1c",
"output_sha256": "dbf7cecb2024c5c42f41bc9c61cfb2b9b8e3592e837b4544dfd659ee80f1eacc"
},
{
"case": "text",
"precision": "q4",
"scale": 2,
"seed": 42,
"steps": 1,
"success": true,
"finite_pixels": true,
"output_size": [
1024,
1024
],
"wall_seconds": 41.21010258299066,
"peak_rss_mb": 1584.88,
"command": [
"/Users/amadeu/Workspace/zimage.swift/zimage.swift/.build/release/ZImageCLI",
"seedvr2",
"--model-path",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4",
"--pos-emb-path",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/pos_emb.safetensors",
"--input",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/examples/text-input.png",
"--output",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/examples/text-q4.png",
"--overwrite",
"--resolution",
"2x",
"--seed",
"42",
"--steps",
"1",
"--tile-size",
"512",
"--tile-overlap",
"64",
"--cache-limit-mb",
"128",
"--memory-limit-mb",
"12000"
],
"binary_sha256": "2d10bb00866722ef9f0525c4dcdde989d29444b6123026b303d507c2259b28dc",
"transformer_sha256": "297e1a8376ab6b96ad381b71d9fc4d7c6c124738e1f07f1b583e688203253086",
"input_sha256": "a480bc72ac400b7140cf50956c0e22054f0ea7f2ed6024edcbee6e60627fab1c",
"output_sha256": "af5b10c72c23f4677a7b766bc12be7e862e9bfa6c6d771ee0a55b4c440a4fa4e",
"versus_fp16": {
"mae": 1.0858135223388672,
"psnr_db": 43.09601841994339
}
},
{
"case": "portrait-4x",
"precision": "fp16",
"scale": 4,
"seed": 42,
"steps": 1,
"success": true,
"finite_pixels": true,
"output_size": [
1024,
1024
],
"wall_seconds": 44.33059312499245,
"peak_rss_mb": 3135.52,
"command": [
"/Users/amadeu/Workspace/zimage.swift/zimage.swift/.build/release/ZImageCLI",
"seedvr2",
"--model-path",
"/Volumes/Z1/seedvr2-1.4b/source",
"--pos-emb-path",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/pos_emb.safetensors",
"--input",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/examples/portrait-4x-input.png",
"--output",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/examples/portrait-4x-fp16.png",
"--overwrite",
"--resolution",
"4x",
"--seed",
"42",
"--steps",
"1",
"--tile-size",
"512",
"--tile-overlap",
"64",
"--cache-limit-mb",
"128",
"--memory-limit-mb",
"12000"
],
"binary_sha256": "2d10bb00866722ef9f0525c4dcdde989d29444b6123026b303d507c2259b28dc",
"transformer_sha256": "58cd0f8e0f0d338ce08255c368dd6befef091f05daa6744cb00b00ba6718e33a",
"input_sha256": "a2bc46bccf4a0f0eb2c4d7e0295a2f2b746da8bd51e571e83d7795b038493811",
"output_sha256": "ef68fce10fa7fb0069d6b98b92e65a67b1b500e0bc90e466b202e54a763c6be3"
},
{
"case": "portrait-4x",
"precision": "q4",
"scale": 4,
"seed": 42,
"steps": 1,
"success": true,
"finite_pixels": true,
"output_size": [
1024,
1024
],
"wall_seconds": 41.23948250000831,
"peak_rss_mb": 1561.28,
"command": [
"/Users/amadeu/Workspace/zimage.swift/zimage.swift/.build/release/ZImageCLI",
"seedvr2",
"--model-path",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4",
"--pos-emb-path",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/pos_emb.safetensors",
"--input",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/examples/portrait-4x-input.png",
"--output",
"/Volumes/Z1/native-model-packages/seedvr2-1.4b-q4/examples/portrait-4x-q4.png",
"--overwrite",
"--resolution",
"4x",
"--seed",
"42",
"--steps",
"1",
"--tile-size",
"512",
"--tile-overlap",
"64",
"--cache-limit-mb",
"128",
"--memory-limit-mb",
"12000"
],
"binary_sha256": "2d10bb00866722ef9f0525c4dcdde989d29444b6123026b303d507c2259b28dc",
"transformer_sha256": "297e1a8376ab6b96ad381b71d9fc4d7c6c124738e1f07f1b583e688203253086",
"input_sha256": "a2bc46bccf4a0f0eb2c4d7e0295a2f2b746da8bd51e571e83d7795b038493811",
"output_sha256": "23e838e8aa97345ad9a1a4729825f1980cabdab6076a84ac909dd001e66ebcad",
"versus_fp16": {
"mae": 1.0141996145248413,
"psnr_db": 43.45973196877463
}
}
],
"note": "Paired native Swift inference; pixel difference is quantization drift, not a quality score. Peak RSS measures host resident memory, not the full GPU allocation. Initial portrait/fox timings overlapped with Swift unit tests and are exploratory."
}