{
 "_about": "Training peak memory: Neurarch plan (shipped factors and held out by architecture) and accelerate estimate-memory, against measured. Generated by scripts/memory-calibration-page.ts; never hand-edit.",
 "factors": {
  "optimizer": 1.2,
  "outputs": 0.88,
  "attention": 3.64,
  "internal": 1.34,
  "recurrent": 5.2,
  "transient": 2.02
 },
 "headroom": 1.15,
 "oomPenalty": 4,
 "cardsGb": [
  16,
  24,
  40,
  80
 ],
 "fittedOn": 178,
 "summary": [
  {
   "key": "heldOut",
   "label": "Neurarch, held out",
   "medianAbsErr": 0.0689,
   "within25": 76,
   "under": 51,
   "saidFitsButOom": 18,
   "saidNoButFits": 19,
   "agree": 665
  },
  {
   "key": "ours",
   "label": "Neurarch, shipped factors",
   "medianAbsErr": 0.0571,
   "within25": 88,
   "under": 52,
   "saidFitsButOom": 15,
   "saidNoButFits": 16,
   "agree": 671
  },
  {
   "key": "accelerateFp32",
   "label": "accelerate, float32",
   "medianAbsErr": 0.4337,
   "within25": 41,
   "under": 97,
   "saidFitsButOom": 123,
   "saidNoButFits": 0,
   "agree": 579
  },
  {
   "key": "accelerateBf16",
   "label": "accelerate, bf16",
   "medianAbsErr": 0.7169,
   "within25": 0,
   "under": 97,
   "saidFitsButOom": 131,
   "saidNoButFits": 0,
   "agree": 571
  }
 ],
 "noFitChecks": 131,
 "oomRuns": 55,
 "regime": {
  "gpu": "L4",
  "factors": [
   1.2017,
   0.9079,
   3.3744,
   1.3214,
   5.1417,
   2.2582
  ],
  "runs": 51,
  "big": 15,
  "medianAbsErr": 0.0992,
  "within25": 13,
  "noFit": 79,
  "flagged": 77,
  "falseNo": 8
 },
 "afterFit": {
  "label": "refit check, 2026-10-17",
  "runs": 25,
  "big": 6,
  "medianAbsErr": 0.14,
  "within25": 5,
  "noFit": 42,
  "flagged": 30,
  "falseNo": 4
 },
 "upperBound": {
  "runs": 91,
  "minGb": 1,
  "shipped": 0.95,
  "bounds": [
   {
    "coverage": 0.9,
    "k": 1.17,
    "noFitFits": 2,
    "noFitBorderline": 27,
    "noFitDoesNot": 102,
    "fitFits": 544,
    "fitBorderline": 14,
    "fitDoesNot": 13,
    "freshNoFitFits": 1,
    "freshCovered": 6,
    "freshMeasured": 6
   },
   {
    "coverage": 0.95,
    "k": 1.43,
    "noFitFits": 0,
    "noFitBorderline": 29,
    "noFitDoesNot": 102,
    "fitFits": 531,
    "fitBorderline": 27,
    "fitDoesNot": 13,
    "freshNoFitFits": 0,
    "freshCovered": 6,
    "freshMeasured": 6
   }
  ]
 },
 "rows": [
  {
   "template": "bert-base",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 32,
   "params": 31120896,
   "measuredPeakGb": 2.2701,
   "heldOutGb": 2.2682,
   "shippedGb": 2.274,
   "accelerateFp32Gb": 0.4979,
   "accelerateBf16Gb": 0.249
  },
  {
   "template": "bst",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 32,
   "params": 128806243,
   "measuredPeakGb": 2.5932,
   "heldOutGb": 2.4594,
   "shippedGb": 2.4778,
   "accelerateFp32Gb": 2.0609,
   "accelerateBf16Gb": 1.0304
  },
  {
   "template": "clip-vit-b32",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 32,
   "params": 151198976,
   "measuredPeakGb": 3.5676,
   "heldOutGb": 3.8448,
   "shippedGb": 3.8517,
   "accelerateFp32Gb": 2.4192,
   "accelerateBf16Gb": 1.2096
  },
  {
   "template": "cnn-lstm-1d",
   "family": "conv",
   "gpu": "A10G",
   "batch": 32,
   "params": 311749,
   "measuredPeakGb": 1.226,
   "heldOutGb": 0.3153,
   "shippedGb": 0.6087,
   "accelerateFp32Gb": 0.005,
   "accelerateBf16Gb": 0.0025
  },
  {
   "template": "colbert",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 32,
   "params": 75430144,
   "measuredPeakGb": 1.5355,
   "heldOutGb": 1.8184,
   "shippedGb": 1.812,
   "accelerateFp32Gb": 1.2069,
   "accelerateBf16Gb": 0.6034
  },
  {
   "template": "din",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 32,
   "params": 128045397,
   "measuredPeakGb": 2.578,
   "heldOutGb": 2.4404,
   "shippedGb": 2.4591,
   "accelerateFp32Gb": 2.0487,
   "accelerateBf16Gb": 1.0244
  },
  {
   "template": "dlrm",
   "family": "other",
   "gpu": "A10G",
   "batch": 32,
   "params": 64349633,
   "measuredPeakGb": 1.304,
   "heldOutGb": 1.2254,
   "shippedGb": 1.2359,
   "accelerateFp32Gb": 1.0296,
   "accelerateBf16Gb": 0.5148
  },
  {
   "template": "dreamer-v3",
   "family": "conv",
   "gpu": "A10G",
   "batch": 32,
   "params": 13711043,
   "measuredPeakGb": 0.2881,
   "heldOutGb": 0.2814,
   "shippedGb": 0.2823,
   "accelerateFp32Gb": 0.2194,
   "accelerateBf16Gb": 0.1097
  },
  {
   "template": "eeg-conformer",
   "family": "vision-transformer",
   "gpu": "A10G",
   "batch": 32,
   "params": 87204,
   "measuredPeakGb": 0.1963,
   "heldOutGb": 0.2101,
   "shippedGb": 0.1913,
   "accelerateFp32Gb": 0.0013,
   "accelerateBf16Gb": 0.0006
  },
  {
   "template": "eegnet",
   "family": "conv",
   "gpu": "A10G",
   "batch": 32,
   "params": 3484,
   "measuredPeakGb": 0.0593,
   "heldOutGb": 0.0454,
   "shippedGb": 0.0466,
   "accelerateFp32Gb": 0,
   "accelerateBf16Gb": 0
  },
  {
   "template": "graph-sage-rec",
   "family": "other",
   "gpu": "A10G",
   "batch": 32,
   "params": 33088,
   "excluded": "the exported example input ([4, 128], [2, 4]) has no batch axis, so the run measured one fixed graph whatever the batch, and the per-sample features do not describe it",
   "measuredPeakGb": 0.0186,
   "heldOutGb": 0.047,
   "shippedGb": 0.0471,
   "accelerateFp32Gb": 0.0042,
   "accelerateBf16Gb": 0.0021
  },
  {
   "template": "ijepa",
   "family": "other",
   "gpu": "A10G",
   "batch": 32,
   "params": 40770432,
   "measuredPeakGb": 1.596,
   "heldOutGb": 1.6613,
   "shippedGb": 1.6708,
   "accelerateFp32Gb": 0.6518,
   "accelerateBf16Gb": 0.3259
  },
  {
   "template": "lightgcn",
   "family": "other",
   "gpu": "A10G",
   "batch": 32,
   "params": 70424960,
   "excluded": "the exported example input ([4], [4], [2, 4]) has no batch axis, so the run measured one fixed graph whatever the batch, and the per-sample features do not describe it",
   "measuredPeakGb": 1.4255,
   "heldOutGb": 1.3487,
   "shippedGb": 1.3522,
   "accelerateFp32Gb": 1.1268,
   "accelerateBf16Gb": 0.5634
  },
  {
   "template": "matryoshka-embed",
   "family": "other",
   "gpu": "A10G",
   "batch": 32,
   "params": 54745345,
   "measuredPeakGb": 3.2392,
   "heldOutGb": 2.5652,
   "shippedGb": 2.7211,
   "accelerateFp32Gb": 0.8759,
   "accelerateBf16Gb": 0.4379
  },
  {
   "template": "mind",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 32,
   "params": 128008256,
   "measuredPeakGb": 2.5772,
   "heldOutGb": 2.4397,
   "shippedGb": 2.4584,
   "accelerateFp32Gb": 2.0481,
   "accelerateBf16Gb": 1.0241
  },
  {
   "template": "ncf",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 32,
   "params": 35206273,
   "measuredPeakGb": 0.7218,
   "heldOutGb": 0.6731,
   "shippedGb": 0.676,
   "accelerateFp32Gb": 0.5633,
   "accelerateBf16Gb": 0.2817
  },
  {
   "template": "neumf",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 32,
   "params": 105610401,
   "measuredPeakGb": 1.9757,
   "heldOutGb": 2.0263,
   "shippedGb": 2.0278,
   "accelerateFp32Gb": 1.6898,
   "accelerateBf16Gb": 0.8449
  },
  {
   "template": "patch-tst",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 32,
   "params": 329860,
   "measuredPeakGb": 0.0816,
   "heldOutGb": 0.0588,
   "shippedGb": 0.0596,
   "accelerateFp32Gb": 0.0063,
   "accelerateBf16Gb": 0.0032
  },
  {
   "template": "resnet-block",
   "family": "conv",
   "gpu": "A10G",
   "batch": 32,
   "params": 74112,
   "measuredPeakGb": 0.0442,
   "heldOutGb": 0.0348,
   "shippedGb": 0.0357,
   "accelerateFp32Gb": 0.0012,
   "accelerateBf16Gb": 0.0006
  },
  {
   "template": "rqvae-semantic-id",
   "family": "other",
   "gpu": "A10G",
   "batch": 32,
   "params": 1247616,
   "measuredPeakGb": 0.0459,
   "heldOutGb": 0.0242,
   "shippedGb": 0.0244,
   "accelerateFp32Gb": 0.02,
   "accelerateBf16Gb": 0.01
  },
  {
   "template": "sim",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 32,
   "params": 128056717,
   "measuredPeakGb": 2.5786,
   "heldOutGb": 2.4667,
   "shippedGb": 2.4827,
   "accelerateFp32Gb": 2.0489,
   "accelerateBf16Gb": 1.0245
  },
  {
   "template": "simple-cnn",
   "family": "conv",
   "gpu": "A10G",
   "batch": 32,
   "params": 804554,
   "measuredPeakGb": 0.0353,
   "heldOutGb": 0.0221,
   "shippedGb": 0.0222,
   "accelerateFp32Gb": 0.0129,
   "accelerateBf16Gb": 0.0064
  },
  {
   "template": "simple-rnn",
   "family": "recurrent",
   "gpu": "A10G",
   "batch": 32,
   "params": 1100298,
   "measuredPeakGb": 0.0911,
   "heldOutGb": 0.2354,
   "shippedGb": 0.1519,
   "accelerateFp32Gb": 0.0176,
   "accelerateBf16Gb": 0.0088
  },
  {
   "template": "sli-rec",
   "family": "recurrent",
   "gpu": "A10G",
   "batch": 32,
   "params": 64054210,
   "measuredPeakGb": 1.2989,
   "heldOutGb": 1.2284,
   "shippedGb": 1.238,
   "accelerateFp32Gb": 1.0248,
   "accelerateBf16Gb": 0.5124
  },
  {
   "template": "t5-small",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 32,
   "params": 56736640,
   "measuredPeakGb": 2.2122,
   "heldOutGb": 2.9238,
   "shippedGb": 2.86,
   "accelerateFp32Gb": 0.9078,
   "accelerateBf16Gb": 0.4539
  },
  {
   "template": "tiger",
   "family": "other",
   "gpu": "A10G",
   "batch": 32,
   "params": 1053824,
   "measuredPeakGb": 0.1216,
   "heldOutGb": 0.0768,
   "shippedGb": 0.0791,
   "accelerateFp32Gb": 0.0169,
   "accelerateBf16Gb": 0.0084
  },
  {
   "template": "transformer-block",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 32,
   "params": 7087872,
   "measuredPeakGb": 1.4506,
   "heldOutGb": 1.3532,
   "shippedGb": 1.3663,
   "accelerateFp32Gb": 0.1134,
   "accelerateBf16Gb": 0.0567
  },
  {
   "template": "two-tower",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 32,
   "params": 70441345,
   "measuredPeakGb": 1.4259,
   "heldOutGb": 1.3412,
   "shippedGb": 1.3525,
   "accelerateFp32Gb": 1.1271,
   "accelerateBf16Gb": 0.5635
  },
  {
   "template": "unet",
   "family": "conv",
   "gpu": "A10G",
   "batch": 32,
   "params": 720705,
   "measuredPeakGb": 6.9695,
   "heldOutGb": 10.5328,
   "shippedGb": 7.3505,
   "accelerateFp32Gb": 0.0115,
   "accelerateBf16Gb": 0.0058
  },
  {
   "template": "vit-b16",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 32,
   "params": 8449000,
   "measuredPeakGb": 0.5764,
   "heldOutGb": 0.5514,
   "shippedGb": 0.5543,
   "accelerateFp32Gb": 0.1352,
   "accelerateBf16Gb": 0.0676
  },
  {
   "template": "wide-and-deep",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 32,
   "params": 3652882,
   "measuredPeakGb": 0.0938,
   "heldOutGb": 0.0701,
   "shippedGb": 0.0706,
   "accelerateFp32Gb": 0.0584,
   "accelerateBf16Gb": 0.0292
  },
  {
   "template": "bert-base",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 8,
   "params": 31120896,
   "measuredPeakGb": 0.885,
   "heldOutGb": 1.0143,
   "shippedGb": 1.0166,
   "accelerateFp32Gb": 0.4979,
   "accelerateBf16Gb": 0.249
  },
  {
   "template": "bst",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 8,
   "params": 128806243,
   "measuredPeakGb": 2.5932,
   "heldOutGb": 2.4559,
   "shippedGb": 2.4742,
   "accelerateFp32Gb": 2.0609,
   "accelerateBf16Gb": 1.0304
  },
  {
   "template": "clip-vit-b32",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 8,
   "params": 151198976,
   "measuredPeakGb": 3.0721,
   "heldOutGb": 3.1378,
   "shippedGb": 3.1402,
   "accelerateFp32Gb": 2.4192,
   "accelerateBf16Gb": 1.2096
  },
  {
   "template": "cnn-lstm-1d",
   "family": "conv",
   "gpu": "A10G",
   "batch": 8,
   "params": 311749,
   "measuredPeakGb": 0.8141,
   "heldOutGb": 0.0833,
   "shippedGb": 0.1567,
   "accelerateFp32Gb": 0.005,
   "accelerateBf16Gb": 0.0025
  },
  {
   "template": "colbert",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 8,
   "params": 75430144,
   "measuredPeakGb": 1.5331,
   "heldOutGb": 1.5437,
   "shippedGb": 1.5392,
   "accelerateFp32Gb": 1.2069,
   "accelerateBf16Gb": 0.6034
  },
  {
   "template": "diffusion-unet",
   "family": "conv",
   "gpu": "A10G",
   "batch": 8,
   "params": 6687044,
   "measuredPeakGb": 17.5151,
   "heldOutGb": 12.3209,
   "shippedGb": 13.1699,
   "accelerateFp32Gb": 0.107,
   "accelerateBf16Gb": 0.0535
  },
  {
   "template": "din",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 8,
   "params": 128045397,
   "measuredPeakGb": 2.578,
   "heldOutGb": 2.4399,
   "shippedGb": 2.4586,
   "accelerateFp32Gb": 2.0487,
   "accelerateBf16Gb": 1.0244
  },
  {
   "template": "dit-xl2",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 8,
   "params": 670684064,
   "measuredPeakGb": 12.179,
   "heldOutGb": 17.4511,
   "shippedGb": 17.1084,
   "accelerateFp32Gb": 10.7309,
   "accelerateBf16Gb": 5.3655
  },
  {
   "template": "dlrm",
   "family": "other",
   "gpu": "A10G",
   "batch": 8,
   "params": 64349633,
   "measuredPeakGb": 1.304,
   "heldOutGb": 1.2251,
   "shippedGb": 1.2356,
   "accelerateFp32Gb": 1.0296,
   "accelerateBf16Gb": 0.5148
  },
  {
   "template": "dreamer-v3",
   "family": "conv",
   "gpu": "A10G",
   "batch": 8,
   "params": 13711043,
   "measuredPeakGb": 0.2821,
   "heldOutGb": 0.2671,
   "shippedGb": 0.268,
   "accelerateFp32Gb": 0.2194,
   "accelerateBf16Gb": 0.1097
  },
  {
   "template": "eeg-conformer",
   "family": "vision-transformer",
   "gpu": "A10G",
   "batch": 8,
   "params": 87204,
   "measuredPeakGb": 0.065,
   "heldOutGb": 0.0537,
   "shippedGb": 0.049,
   "accelerateFp32Gb": 0.0013,
   "accelerateBf16Gb": 0.0006
  },
  {
   "template": "eegnet",
   "family": "conv",
   "gpu": "A10G",
   "batch": 8,
   "params": 3484,
   "measuredPeakGb": 0.0293,
   "heldOutGb": 0.0114,
   "shippedGb": 0.0117,
   "accelerateFp32Gb": 0,
   "accelerateBf16Gb": 0
  },
  {
   "template": "gpt2",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 8,
   "params": 84334417,
   "measuredPeakGb": 5.2491,
   "heldOutGb": 4.9401,
   "shippedGb": 5.0097,
   "accelerateFp32Gb": 1.3493,
   "accelerateBf16Gb": 0.6747
  },
  {
   "template": "graph-sage-rec",
   "family": "other",
   "gpu": "A10G",
   "batch": 8,
   "params": 33088,
   "excluded": "the exported example input ([4, 128], [2, 4]) has no batch axis, so the run measured one fixed graph whatever the batch, and the per-sample features do not describe it",
   "measuredPeakGb": 0.0186,
   "heldOutGb": 0.0155,
   "shippedGb": 0.0156,
   "accelerateFp32Gb": 0.0042,
   "accelerateBf16Gb": 0.0021
  },
  {
   "template": "ijepa",
   "family": "other",
   "gpu": "A10G",
   "batch": 8,
   "params": 40770432,
   "measuredPeakGb": 0.7357,
   "heldOutGb": 1.0022,
   "shippedGb": 1.0044,
   "accelerateFp32Gb": 0.6518,
   "accelerateBf16Gb": 0.3259
  },
  {
   "template": "lightgcn",
   "family": "other",
   "gpu": "A10G",
   "batch": 8,
   "params": 70424960,
   "excluded": "the exported example input ([4], [4], [2, 4]) has no batch axis, so the run measured one fixed graph whatever the batch, and the per-sample features do not describe it",
   "measuredPeakGb": 1.4255,
   "heldOutGb": 1.3487,
   "shippedGb": 1.3522,
   "accelerateFp32Gb": 1.1268,
   "accelerateBf16Gb": 0.5634
  },
  {
   "template": "llama3-block",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 8,
   "params": 702554112,
   "measuredPeakGb": 14.3367,
   "heldOutGb": 16.064,
   "shippedGb": 16.1269,
   "accelerateFp32Gb": 11.2409,
   "accelerateBf16Gb": 5.6204
  },
  {
   "template": "matryoshka-embed",
   "family": "other",
   "gpu": "A10G",
   "batch": 8,
   "params": 54745345,
   "measuredPeakGb": 1.3696,
   "heldOutGb": 1.4334,
   "shippedGb": 1.4686,
   "accelerateFp32Gb": 0.8759,
   "accelerateBf16Gb": 0.4379
  },
  {
   "template": "mind",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 8,
   "params": 128008256,
   "measuredPeakGb": 2.5772,
   "heldOutGb": 2.4392,
   "shippedGb": 2.4579,
   "accelerateFp32Gb": 2.0481,
   "accelerateBf16Gb": 1.0241
  },
  {
   "template": "ncf",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 8,
   "params": 35206273,
   "measuredPeakGb": 0.7212,
   "heldOutGb": 0.6731,
   "shippedGb": 0.676,
   "accelerateFp32Gb": 0.5633,
   "accelerateBf16Gb": 0.2817
  },
  {
   "template": "neumf",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 8,
   "params": 105610401,
   "measuredPeakGb": 1.9757,
   "heldOutGb": 2.0262,
   "shippedGb": 2.0277,
   "accelerateFp32Gb": 1.6898,
   "accelerateBf16Gb": 0.8449
  },
  {
   "template": "patch-tst",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 8,
   "params": 329860,
   "measuredPeakGb": 0.0376,
   "heldOutGb": 0.0204,
   "shippedGb": 0.0206,
   "accelerateFp32Gb": 0.0063,
   "accelerateBf16Gb": 0.0032
  },
  {
   "template": "resnet-block",
   "family": "conv",
   "gpu": "A10G",
   "batch": 8,
   "params": 74112,
   "measuredPeakGb": 0.0296,
   "heldOutGb": 0.0098,
   "shippedGb": 0.01,
   "accelerateFp32Gb": 0.0012,
   "accelerateBf16Gb": 0.0006
  },
  {
   "template": "rqvae-semantic-id",
   "family": "other",
   "gpu": "A10G",
   "batch": 8,
   "params": 1247616,
   "measuredPeakGb": 0.0414,
   "heldOutGb": 0.0239,
   "shippedGb": 0.0241,
   "accelerateFp32Gb": 0.02,
   "accelerateBf16Gb": 0.01
  },
  {
   "template": "sim",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 8,
   "params": 128056717,
   "measuredPeakGb": 2.5783,
   "heldOutGb": 2.4487,
   "shippedGb": 2.4647,
   "accelerateFp32Gb": 2.0489,
   "accelerateBf16Gb": 1.0245
  },
  {
   "template": "simple-cnn",
   "family": "conv",
   "gpu": "A10G",
   "batch": 8,
   "params": 804554,
   "measuredPeakGb": 0.0335,
   "heldOutGb": 0.017,
   "shippedGb": 0.0171,
   "accelerateFp32Gb": 0.0129,
   "accelerateBf16Gb": 0.0064
  },
  {
   "template": "simple-rnn",
   "family": "recurrent",
   "gpu": "A10G",
   "batch": 8,
   "params": 1100298,
   "measuredPeakGb": 0.0502,
   "heldOutGb": 0.0746,
   "shippedGb": 0.0538,
   "accelerateFp32Gb": 0.0176,
   "accelerateBf16Gb": 0.0088
  },
  {
   "template": "sli-rec",
   "family": "recurrent",
   "gpu": "A10G",
   "batch": 8,
   "params": 64054210,
   "measuredPeakGb": 1.2981,
   "heldOutGb": 1.2222,
   "shippedGb": 1.2318,
   "accelerateFp32Gb": 1.0248,
   "accelerateBf16Gb": 0.5124
  },
  {
   "template": "swin-tiny",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 8,
   "params": 28260808,
   "measuredPeakGb": 5.7964,
   "heldOutGb": 6.0812,
   "shippedGb": 5.9757,
   "accelerateFp32Gb": 0.4522,
   "accelerateBf16Gb": 0.2261
  },
  {
   "template": "t5-small",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 8,
   "params": 56736640,
   "measuredPeakGb": 1.0007,
   "heldOutGb": 1.5593,
   "shippedGb": 1.532,
   "accelerateFp32Gb": 0.9078,
   "accelerateBf16Gb": 0.4539
  },
  {
   "template": "tiger",
   "family": "other",
   "gpu": "A10G",
   "batch": 8,
   "params": 1053824,
   "measuredPeakGb": 0.0545,
   "heldOutGb": 0.0344,
   "shippedGb": 0.0349,
   "accelerateFp32Gb": 0.0169,
   "accelerateBf16Gb": 0.0084
  },
  {
   "template": "transformer-block",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 8,
   "params": 7087872,
   "measuredPeakGb": 0.4579,
   "heldOutGb": 0.4402,
   "shippedGb": 0.4436,
   "accelerateFp32Gb": 0.1134,
   "accelerateBf16Gb": 0.0567
  },
  {
   "template": "two-tower",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 8,
   "params": 70441345,
   "measuredPeakGb": 1.4259,
   "heldOutGb": 1.3411,
   "shippedGb": 1.3525,
   "accelerateFp32Gb": 1.1271,
   "accelerateBf16Gb": 0.5635
  },
  {
   "template": "unet",
   "family": "conv",
   "gpu": "A10G",
   "batch": 8,
   "params": 720705,
   "measuredPeakGb": 1.7625,
   "heldOutGb": 2.6435,
   "shippedGb": 1.848,
   "accelerateFp32Gb": 0.0115,
   "accelerateBf16Gb": 0.0058
  },
  {
   "template": "video-dit",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 8,
   "params": 99434192,
   "measuredPeakGb": 5.0443,
   "heldOutGb": 5.4386,
   "shippedGb": 5.4197,
   "accelerateFp32Gb": 1.5889,
   "accelerateBf16Gb": 0.7944
  },
  {
   "template": "vit-b16",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 8,
   "params": 8449000,
   "measuredPeakGb": 0.2651,
   "heldOutGb": 0.2593,
   "shippedGb": 0.2602,
   "accelerateFp32Gb": 0.1352,
   "accelerateBf16Gb": 0.0676
  },
  {
   "template": "wide-and-deep",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 8,
   "params": 3652882,
   "measuredPeakGb": 0.0918,
   "heldOutGb": 0.0697,
   "shippedGb": 0.0702,
   "accelerateFp32Gb": 0.0584,
   "accelerateBf16Gb": 0.0292
  },
  {
   "template": "bert-base",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 96,
   "params": 31120896,
   "measuredPeakGb": 5.9688,
   "heldOutGb": 5.612,
   "shippedGb": 5.627,
   "accelerateFp32Gb": 0.4979,
   "accelerateBf16Gb": 0.249
  },
  {
   "template": "bst",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 96,
   "params": 128806243,
   "measuredPeakGb": 2.5939,
   "heldOutGb": 2.4688,
   "shippedGb": 2.4871,
   "accelerateFp32Gb": 2.0609,
   "accelerateBf16Gb": 1.0304
  },
  {
   "template": "clip-vit-b32",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 96,
   "params": 151198976,
   "measuredPeakGb": 6.3584,
   "heldOutGb": 5.73,
   "shippedGb": 5.749,
   "accelerateFp32Gb": 2.4192,
   "accelerateBf16Gb": 1.2096
  },
  {
   "template": "cnn-lstm-1d",
   "family": "conv",
   "gpu": "A10G",
   "batch": 96,
   "params": 311749,
   "measuredPeakGb": 2.3132,
   "heldOutGb": 0.9339,
   "shippedGb": 1.8141,
   "accelerateFp32Gb": 0.005,
   "accelerateBf16Gb": 0.0025
  },
  {
   "template": "colbert",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 96,
   "params": 75430144,
   "measuredPeakGb": 2.5494,
   "heldOutGb": 2.5508,
   "shippedGb": 2.5395,
   "accelerateFp32Gb": 1.2069,
   "accelerateBf16Gb": 0.6034
  },
  {
   "template": "din",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 96,
   "params": 128045397,
   "measuredPeakGb": 2.578,
   "heldOutGb": 2.4417,
   "shippedGb": 2.4604,
   "accelerateFp32Gb": 2.0487,
   "accelerateBf16Gb": 1.0244
  },
  {
   "template": "dlrm",
   "family": "other",
   "gpu": "A10G",
   "batch": 96,
   "params": 64349633,
   "measuredPeakGb": 1.3041,
   "heldOutGb": 1.2263,
   "shippedGb": 1.2368,
   "accelerateFp32Gb": 1.0296,
   "accelerateBf16Gb": 0.5148
  },
  {
   "template": "dreamer-v3",
   "family": "conv",
   "gpu": "A10G",
   "batch": 96,
   "params": 13711043,
   "measuredPeakGb": 0.2899,
   "heldOutGb": 0.3195,
   "shippedGb": 0.3205,
   "accelerateFp32Gb": 0.2194,
   "accelerateBf16Gb": 0.1097
  },
  {
   "template": "eeg-conformer",
   "family": "vision-transformer",
   "gpu": "A10G",
   "batch": 96,
   "params": 87204,
   "measuredPeakGb": 0.544,
   "heldOutGb": 0.6273,
   "shippedGb": 0.571,
   "accelerateFp32Gb": 0.0013,
   "accelerateBf16Gb": 0.0006
  },
  {
   "template": "eegnet",
   "family": "conv",
   "gpu": "A10G",
   "batch": 96,
   "params": 3484,
   "measuredPeakGb": 0.1343,
   "heldOutGb": 0.1362,
   "shippedGb": 0.1397,
   "accelerateFp32Gb": 0,
   "accelerateBf16Gb": 0
  },
  {
   "template": "graph-sage-rec",
   "family": "other",
   "gpu": "A10G",
   "batch": 96,
   "params": 33088,
   "excluded": "the exported example input ([4, 128], [2, 4]) has no batch axis, so the run measured one fixed graph whatever the batch, and the per-sample features do not describe it",
   "measuredPeakGb": 0.0186,
   "heldOutGb": 0.1309,
   "shippedGb": 0.1313,
   "accelerateFp32Gb": 0.0042,
   "accelerateBf16Gb": 0.0021
  },
  {
   "template": "ijepa",
   "family": "other",
   "gpu": "A10G",
   "batch": 96,
   "params": 40770432,
   "measuredPeakGb": 3.8626,
   "heldOutGb": 3.419,
   "shippedGb": 3.4479,
   "accelerateFp32Gb": 0.6518,
   "accelerateBf16Gb": 0.3259
  },
  {
   "template": "lightgcn",
   "family": "other",
   "gpu": "A10G",
   "batch": 96,
   "params": 70424960,
   "excluded": "the exported example input ([4], [4], [2, 4]) has no batch axis, so the run measured one fixed graph whatever the batch, and the per-sample features do not describe it",
   "measuredPeakGb": 1.4255,
   "heldOutGb": 1.3488,
   "shippedGb": 1.3523,
   "accelerateFp32Gb": 1.1268,
   "accelerateBf16Gb": 0.5634
  },
  {
   "template": "matryoshka-embed",
   "family": "other",
   "gpu": "A10G",
   "batch": 96,
   "params": 54745345,
   "measuredPeakGb": 8.2258,
   "heldOutGb": 5.5834,
   "shippedGb": 6.0613,
   "accelerateFp32Gb": 0.8759,
   "accelerateBf16Gb": 0.4379
  },
  {
   "template": "mind",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 96,
   "params": 128008256,
   "measuredPeakGb": 2.5776,
   "heldOutGb": 2.4409,
   "shippedGb": 2.4597,
   "accelerateFp32Gb": 2.0481,
   "accelerateBf16Gb": 1.0241
  },
  {
   "template": "ncf",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 96,
   "params": 35206273,
   "measuredPeakGb": 0.7212,
   "heldOutGb": 0.6731,
   "shippedGb": 0.676,
   "accelerateFp32Gb": 0.5633,
   "accelerateBf16Gb": 0.2817
  },
  {
   "template": "neumf",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 96,
   "params": 105610401,
   "measuredPeakGb": 1.9757,
   "heldOutGb": 2.0264,
   "shippedGb": 2.0279,
   "accelerateFp32Gb": 1.6898,
   "accelerateBf16Gb": 0.8449
  },
  {
   "template": "patch-tst",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 96,
   "params": 329860,
   "measuredPeakGb": 0.2005,
   "heldOutGb": 0.1611,
   "shippedGb": 0.1636,
   "accelerateFp32Gb": 0.0063,
   "accelerateBf16Gb": 0.0032
  },
  {
   "template": "resnet-block",
   "family": "conv",
   "gpu": "A10G",
   "batch": 96,
   "params": 74112,
   "measuredPeakGb": 0.1451,
   "heldOutGb": 0.1015,
   "shippedGb": 0.1043,
   "accelerateFp32Gb": 0.0012,
   "accelerateBf16Gb": 0.0006
  },
  {
   "template": "rqvae-semantic-id",
   "family": "other",
   "gpu": "A10G",
   "batch": 96,
   "params": 1247616,
   "measuredPeakGb": 0.0714,
   "heldOutGb": 0.025,
   "shippedGb": 0.0252,
   "accelerateFp32Gb": 0.02,
   "accelerateBf16Gb": 0.01
  },
  {
   "template": "sim",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 96,
   "params": 128056717,
   "measuredPeakGb": 2.5797,
   "heldOutGb": 2.5146,
   "shippedGb": 2.5307,
   "accelerateFp32Gb": 2.0489,
   "accelerateBf16Gb": 1.0245
  },
  {
   "template": "simple-cnn",
   "family": "conv",
   "gpu": "A10G",
   "batch": 96,
   "params": 804554,
   "measuredPeakGb": 0.0468,
   "heldOutGb": 0.0355,
   "shippedGb": 0.0358,
   "accelerateFp32Gb": 0.0129,
   "accelerateBf16Gb": 0.0064
  },
  {
   "template": "simple-rnn",
   "family": "recurrent",
   "gpu": "A10G",
   "batch": 96,
   "params": 1100298,
   "measuredPeakGb": 0.2026,
   "heldOutGb": 0.6642,
   "shippedGb": 0.4134,
   "accelerateFp32Gb": 0.0176,
   "accelerateBf16Gb": 0.0088
  },
  {
   "template": "sli-rec",
   "family": "recurrent",
   "gpu": "A10G",
   "batch": 96,
   "params": 64054210,
   "measuredPeakGb": 1.2982,
   "heldOutGb": 1.245,
   "shippedGb": 1.2546,
   "accelerateFp32Gb": 1.0248,
   "accelerateBf16Gb": 0.5124
  },
  {
   "template": "t5-small",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 96,
   "params": 56736640,
   "measuredPeakGb": 5.4462,
   "heldOutGb": 6.5624,
   "shippedGb": 6.4013,
   "accelerateFp32Gb": 0.9078,
   "accelerateBf16Gb": 0.4539
  },
  {
   "template": "tiger",
   "family": "other",
   "gpu": "A10G",
   "batch": 96,
   "params": 1053824,
   "measuredPeakGb": 0.2958,
   "heldOutGb": 0.19,
   "shippedGb": 0.1967,
   "accelerateFp32Gb": 0.0169,
   "accelerateBf16Gb": 0.0084
  },
  {
   "template": "transformer-block",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 96,
   "params": 7087872,
   "measuredPeakGb": 4.1002,
   "heldOutGb": 3.7877,
   "shippedGb": 3.8266,
   "accelerateFp32Gb": 0.1134,
   "accelerateBf16Gb": 0.0567
  },
  {
   "template": "two-tower",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 96,
   "params": 70441345,
   "measuredPeakGb": 1.4259,
   "heldOutGb": 1.3413,
   "shippedGb": 1.3527,
   "accelerateFp32Gb": 1.1271,
   "accelerateBf16Gb": 0.5635
  },
  {
   "template": "vit-b16",
   "family": "transformer",
   "gpu": "A10G",
   "batch": 96,
   "params": 8449000,
   "measuredPeakGb": 1.45,
   "heldOutGb": 1.3305,
   "shippedGb": 1.3384,
   "accelerateFp32Gb": 0.1352,
   "accelerateBf16Gb": 0.0676
  },
  {
   "template": "wide-and-deep",
   "family": "embedding-mlp",
   "gpu": "A10G",
   "batch": 96,
   "params": 3652882,
   "measuredPeakGb": 0.0944,
   "heldOutGb": 0.0709,
   "shippedGb": 0.0714,
   "accelerateFp32Gb": 0.0584,
   "accelerateBf16Gb": 0.0292
  },
  {
   "template": "bert-base",
   "family": "transformer",
   "gpu": "H100",
   "batch": 32,
   "params": 31120896,
   "measuredPeakGb": 2.3203,
   "heldOutGb": 2.2682,
   "shippedGb": 2.274,
   "accelerateFp32Gb": 0.4979,
   "accelerateBf16Gb": 0.249
  },
  {
   "template": "bst",
   "family": "embedding-mlp",
   "gpu": "H100",
   "batch": 32,
   "params": 128806243,
   "measuredPeakGb": 2.6433,
   "heldOutGb": 2.4594,
   "shippedGb": 2.4778,
   "accelerateFp32Gb": 2.0609,
   "accelerateBf16Gb": 1.0304
  },
  {
   "template": "clip-vit-b32",
   "family": "embedding-mlp",
   "gpu": "H100",
   "batch": 32,
   "params": 151198976,
   "measuredPeakGb": 3.5968,
   "heldOutGb": 3.8448,
   "shippedGb": 3.8517,
   "accelerateFp32Gb": 2.4192,
   "accelerateBf16Gb": 1.2096
  },
  {
   "template": "cnn-lstm-1d",
   "family": "conv",
   "gpu": "H100",
   "batch": 32,
   "params": 311749,
   "measuredPeakGb": 1.2711,
   "heldOutGb": 0.3153,
   "shippedGb": 0.6087,
   "accelerateFp32Gb": 0.005,
   "accelerateBf16Gb": 0.0025
  },
  {
   "template": "colbert",
   "family": "embedding-mlp",
   "gpu": "H100",
   "batch": 32,
   "params": 75430144,
   "measuredPeakGb": 1.5873,
   "heldOutGb": 1.8184,
   "shippedGb": 1.812,
   "accelerateFp32Gb": 1.2069,
   "accelerateBf16Gb": 0.6034
  },
  {
   "template": "din",
   "family": "embedding-mlp",
   "gpu": "H100",
   "batch": 32,
   "params": 128045397,
   "measuredPeakGb": 2.6281,
   "heldOutGb": 2.4404,
   "shippedGb": 2.4591,
   "accelerateFp32Gb": 2.0487,
   "accelerateBf16Gb": 1.0244
  },
  {
   "template": "dit-xl2",
   "family": "transformer",
   "gpu": "H100",
   "batch": 32,
   "params": 670684064,
   "measuredPeakGb": 25.6225,
   "heldOutGb": 30.7987,
   "shippedGb": 29.8023,
   "accelerateFp32Gb": 10.7309,
   "accelerateBf16Gb": 5.3655
  },
  {
   "template": "dlrm",
   "family": "other",
   "gpu": "H100",
   "batch": 32,
   "params": 64349633,
   "measuredPeakGb": 1.3541,
   "heldOutGb": 1.2254,
   "shippedGb": 1.2359,
   "accelerateFp32Gb": 1.0296,
   "accelerateBf16Gb": 0.5148
  },
  {
   "template": "dreamer-v3",
   "family": "conv",
   "gpu": "H100",
   "batch": 32,
   "params": 13711043,
   "measuredPeakGb": 0.3363,
   "heldOutGb": 0.2814,
   "shippedGb": 0.2823,
   "accelerateFp32Gb": 0.2194,
   "accelerateBf16Gb": 0.1097
  },
  {
   "template": "eeg-conformer",
   "family": "vision-transformer",
   "gpu": "H100",
   "batch": 32,
   "params": 87204,
   "measuredPeakGb": 0.2477,
   "heldOutGb": 0.2101,
   "shippedGb": 0.1913,
   "accelerateFp32Gb": 0.0013,
   "accelerateBf16Gb": 0.0006
  },
  {
   "template": "eegnet",
   "family": "conv",
   "gpu": "H100",
   "batch": 32,
   "params": 3484,
   "measuredPeakGb": 0.1063,
   "heldOutGb": 0.0454,
   "shippedGb": 0.0466,
   "accelerateFp32Gb": 0,
   "accelerateBf16Gb": 0
  },
  {
   "template": "gpt2",
   "family": "embedding-mlp",
   "gpu": "H100",
   "batch": 32,
   "params": 84334417,
   "measuredPeakGb": 17.6715,
   "heldOutGb": 14.9178,
   "shippedGb": 15.1813,
   "accelerateFp32Gb": 1.3493,
   "accelerateBf16Gb": 0.6747
  },
  {
   "template": "graph-sage-rec",
   "family": "other",
   "gpu": "H100",
   "batch": 32,
   "params": 33088,
   "excluded": "the exported example input ([4, 128], [2, 4]) has no batch axis, so the run measured one fixed graph whatever the batch, and the per-sample features do not describe it",
   "measuredPeakGb": 0.0686,
   "heldOutGb": 0.047,
   "shippedGb": 0.0471,
   "accelerateFp32Gb": 0.0042,
   "accelerateBf16Gb": 0.0021
  },
  {
   "template": "ijepa",
   "family": "other",
   "gpu": "H100",
   "batch": 32,
   "params": 40770432,
   "measuredPeakGb": 1.6376,
   "heldOutGb": 1.6613,
   "shippedGb": 1.6708,
   "accelerateFp32Gb": 0.6518,
   "accelerateBf16Gb": 0.3259
  },
  {
   "template": "lightgcn",
   "family": "other",
   "gpu": "H100",
   "batch": 32,
   "params": 70424960,
   "excluded": "the exported example input ([4], [4], [2, 4]) has no batch axis, so the run measured one fixed graph whatever the batch, and the per-sample features do not describe it",
   "measuredPeakGb": 1.4756,
   "heldOutGb": 1.3487,
   "shippedGb": 1.3522,
   "accelerateFp32Gb": 1.1268,
   "accelerateBf16Gb": 0.5634
  },
  {
   "template": "llama3-block",
   "family": "transformer",
   "gpu": "H100",
   "batch": 32,
   "params": 702554112,
   "measuredPeakGb": 29.1285,
   "heldOutGb": 23.8705,
   "shippedGb": 24.0406,
   "accelerateFp32Gb": 11.2409,
   "accelerateBf16Gb": 5.6204
  },
  {
   "template": "matryoshka-embed",
   "family": "other",
   "gpu": "H100",
   "batch": 32,
   "params": 54745345,
   "measuredPeakGb": 3.2873,
   "heldOutGb": 2.5652,
   "shippedGb": 2.7211,
   "accelerateFp32Gb": 0.8759,
   "accelerateBf16Gb": 0.4379
  },
  {
   "template": "mind",
   "family": "embedding-mlp",
   "gpu": "H100",
   "batch": 32,
   "params": 128008256,
   "measuredPeakGb": 2.6273,
   "heldOutGb": 2.4397,
   "shippedGb": 2.4584,
   "accelerateFp32Gb": 2.0481,
   "accelerateBf16Gb": 1.0241
  },
  {
   "template": "ncf",
   "family": "embedding-mlp",
   "gpu": "H100",
   "batch": 32,
   "params": 35206273,
   "measuredPeakGb": 0.7712,
   "heldOutGb": 0.6731,
   "shippedGb": 0.676,
   "accelerateFp32Gb": 0.5633,
   "accelerateBf16Gb": 0.2817
  },
  {
   "template": "neumf",
   "family": "embedding-mlp",
   "gpu": "H100",
   "batch": 32,
   "params": 105610401,
   "measuredPeakGb": 2.0257,
   "heldOutGb": 2.0263,
   "shippedGb": 2.0278,
   "accelerateFp32Gb": 1.6898,
   "accelerateBf16Gb": 0.8449
  },
  {
   "template": "patch-tst",
   "family": "transformer",
   "gpu": "H100",
   "batch": 32,
   "params": 329860,
   "measuredPeakGb": 0.1307,
   "heldOutGb": 0.0588,
   "shippedGb": 0.0596,
   "accelerateFp32Gb": 0.0063,
   "accelerateBf16Gb": 0.0032
  },
  {
   "template": "qwen2-vl",
   "family": "other",
   "gpu": "H100",
   "batch": 32,
   "params": 1549126912,
   "measuredPeakGb": 41.1112,
   "heldOutGb": 45.2844,
   "shippedGb": 45.2764,
   "accelerateFp32Gb": 24.7862,
   "accelerateBf16Gb": 12.3931
  },
  {
   "template": "resnet-block",
   "family": "conv",
   "gpu": "H100",
   "batch": 32,
   "params": 74112,
   "measuredPeakGb": 0.1122,
   "heldOutGb": 0.0348,
   "shippedGb": 0.0357,
   "accelerateFp32Gb": 0.0012,
   "accelerateBf16Gb": 0.0006
  },
  {
   "template": "rqvae-semantic-id",
   "family": "other",
   "gpu": "H100",
   "batch": 32,
   "params": 1247616,
   "measuredPeakGb": 0.0952,
   "heldOutGb": 0.0242,
   "shippedGb": 0.0244,
   "accelerateFp32Gb": 0.02,
   "accelerateBf16Gb": 0.01
  },
  {
   "template": "sim",
   "family": "embedding-mlp",
   "gpu": "H100",
   "batch": 32,
   "params": 128056717,
   "measuredPeakGb": 2.6287,
   "heldOutGb": 2.4667,
   "shippedGb": 2.4827,
   "accelerateFp32Gb": 2.0489,
   "accelerateBf16Gb": 1.0245
  },
  {
   "template": "simple-cnn",
   "family": "conv",
   "gpu": "H100",
   "batch": 32,
   "params": 804554,
   "measuredPeakGb": 0.0892,
   "heldOutGb": 0.0221,
   "shippedGb": 0.0222,
   "accelerateFp32Gb": 0.0129,
   "accelerateBf16Gb": 0.0064
  },
  {
   "template": "simple-rnn",
   "family": "recurrent",
   "gpu": "H100",
   "batch": 32,
   "params": 1100298,
   "measuredPeakGb": 0.1577,
   "heldOutGb": 0.2354,
   "shippedGb": 0.1519,
   "accelerateFp32Gb": 0.0176,
   "accelerateBf16Gb": 0.0088
  },
  {
   "template": "sli-rec",
   "family": "recurrent",
   "gpu": "H100",
   "batch": 32,
   "params": 64054210,
   "measuredPeakGb": 1.3482,
   "heldOutGb": 1.2284,
   "shippedGb": 1.238,
   "accelerateFp32Gb": 1.0248,
   "accelerateBf16Gb": 0.5124
  },
  {
   "template": "swin-tiny",
   "family": "transformer",
   "gpu": "H100",
   "batch": 32,
   "params": 28260808,
   "measuredPeakGb": 21.7945,
   "heldOutGb": 22.7015,
   "shippedGb": 22.2751,
   "accelerateFp32Gb": 0.4522,
   "accelerateBf16Gb": 0.2261
  },
  {
   "template": "t5-small",
   "family": "transformer",
   "gpu": "H100",
   "batch": 32,
   "params": 56736640,
   "measuredPeakGb": 2.2594,
   "heldOutGb": 2.9238,
   "shippedGb": 2.86,
   "accelerateFp32Gb": 0.9078,
   "accelerateBf16Gb": 0.4539
  },
  {
   "template": "tiger",
   "family": "other",
   "gpu": "H100",
   "batch": 32,
   "params": 1053824,
   "measuredPeakGb": 0.1701,
   "heldOutGb": 0.0768,
   "shippedGb": 0.0791,
   "accelerateFp32Gb": 0.0169,
   "accelerateBf16Gb": 0.0084
  },
  {
   "template": "transformer-block",
   "family": "transformer",
   "gpu": "H100",
   "batch": 32,
   "params": 7087872,
   "measuredPeakGb": 1.4996,
   "heldOutGb": 1.3532,
   "shippedGb": 1.3663,
   "accelerateFp32Gb": 0.1134,
   "accelerateBf16Gb": 0.0567
  },
  {
   "template": "two-tower",
   "family": "embedding-mlp",
   "gpu": "H100",
   "batch": 32,
   "params": 70441345,
   "measuredPeakGb": 1.4759,
   "heldOutGb": 1.3412,
   "shippedGb": 1.3525,
   "accelerateFp32Gb": 1.1271,
   "accelerateBf16Gb": 0.5635
  },
  {
   "template": "unet",
   "family": "conv",
   "gpu": "H100",
   "batch": 32,
   "params": 720705,
   "measuredPeakGb": 7.0197,
   "heldOutGb": 10.5328,
   "shippedGb": 7.3505,
   "accelerateFp32Gb": 0.0115,
   "accelerateBf16Gb": 0.0058
  },
  {
   "template": "video-dit",
   "family": "transformer",
   "gpu": "H100",
   "batch": 32,
   "params": 99434192,
   "measuredPeakGb": 15.9642,
   "heldOutGb": 16.0515,
   "shippedGb": 15.9587,
   "accelerateFp32Gb": 1.5889,
   "accelerateBf16Gb": 0.7944
  },
  {
   "template": "vit-b16",
   "family": "transformer",
   "gpu": "H100",
   "batch": 32,
   "params": 8449000,
   "measuredPeakGb": 0.6367,
   "heldOutGb": 0.5514,
   "shippedGb": 0.5543,
   "accelerateFp32Gb": 0.1352,
   "accelerateBf16Gb": 0.0676
  },
  {
   "template": "wide-and-deep",
   "family": "embedding-mlp",
   "gpu": "H100",
   "batch": 32,
   "params": 3652882,
   "measuredPeakGb": 0.1418,
   "heldOutGb": 0.0701,
   "shippedGb": 0.0706,
   "accelerateFp32Gb": 0.0584,
   "accelerateBf16Gb": 0.0292
  },
  {
   "template": "bert-base",
   "family": "transformer",
   "gpu": "L4",
   "batch": 384,
   "params": 31120896,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 20.6587,
   "shippedGb": 20.7153,
   "accelerateFp32Gb": 0.4979,
   "accelerateBf16Gb": 0.249
  },
  {
   "template": "bert-base",
   "family": "transformer",
   "gpu": "L4",
   "batch": 448,
   "params": 31120896,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 24.0025,
   "shippedGb": 24.0683,
   "accelerateFp32Gb": 0.4979,
   "accelerateBf16Gb": 0.249
  },
  {
   "template": "bert-base",
   "family": "transformer",
   "gpu": "L4",
   "batch": 512,
   "params": 31120896,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 27.3462,
   "shippedGb": 27.4212,
   "accelerateFp32Gb": 0.4979,
   "accelerateBf16Gb": 0.249
  },
  {
   "template": "bert-base",
   "family": "transformer",
   "gpu": "L4",
   "batch": 768,
   "params": 31120896,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 40.7211,
   "shippedGb": 40.8331,
   "accelerateFp32Gb": 0.4979,
   "accelerateBf16Gb": 0.249
  },
  {
   "template": "clip-vit-b32",
   "family": "embedding-mlp",
   "gpu": "L4",
   "batch": 512,
   "params": 151198976,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 17.9841,
   "shippedGb": 18.0818,
   "accelerateFp32Gb": 2.4192,
   "accelerateBf16Gb": 1.2096
  },
  {
   "template": "clip-vit-b32",
   "family": "embedding-mlp",
   "gpu": "L4",
   "batch": 640,
   "params": 151198976,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 21.7545,
   "shippedGb": 21.8765,
   "accelerateFp32Gb": 2.4192,
   "accelerateBf16Gb": 1.2096
  },
  {
   "template": "clip-vit-b32",
   "family": "embedding-mlp",
   "gpu": "L4",
   "batch": 4096,
   "params": 151198976,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 123.5574,
   "shippedGb": 124.3336,
   "accelerateFp32Gb": 2.4192,
   "accelerateBf16Gb": 1.2096
  },
  {
   "template": "cnn-lstm-1d",
   "family": "conv",
   "gpu": "L4",
   "batch": 1024,
   "params": 311749,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": 18.0927,
   "heldOutGb": 9.9043,
   "shippedGb": 19.2925,
   "accelerateFp32Gb": 0.005,
   "accelerateBf16Gb": 0.0025
  },
  {
   "template": "cnn-lstm-1d",
   "family": "conv",
   "gpu": "L4",
   "batch": 2048,
   "params": 311749,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 19.8027,
   "shippedGb": 38.5789,
   "accelerateFp32Gb": 0.005,
   "accelerateBf16Gb": 0.0025
  },
  {
   "template": "cnn-lstm-1d",
   "family": "conv",
   "gpu": "L4",
   "batch": 4096,
   "params": 311749,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 39.5995,
   "shippedGb": 77.1519,
   "accelerateFp32Gb": 0.005,
   "accelerateBf16Gb": 0.0025
  },
  {
   "template": "colbert",
   "family": "embedding-mlp",
   "gpu": "L4",
   "batch": 1536,
   "params": 75430144,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 19.0305,
   "shippedGb": 18.9079,
   "accelerateFp32Gb": 1.2069,
   "accelerateBf16Gb": 0.6034
  },
  {
   "template": "colbert",
   "family": "embedding-mlp",
   "gpu": "L4",
   "batch": 2048,
   "params": 75430144,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 24.8899,
   "shippedGb": 24.7278,
   "accelerateFp32Gb": 1.2069,
   "accelerateBf16Gb": 0.6034
  },
  {
   "template": "deepseek-v3-r",
   "family": "transformer",
   "gpu": "L4",
   "batch": 1,
   "params": 1080932352,
   "variantOf": "deepseek-v3",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 22.3629,
   "shippedGb": 22.4179,
   "accelerateFp32Gb": 17.2614,
   "accelerateBf16Gb": 8.6307
  },
  {
   "template": "deepseek-v3-r",
   "family": "transformer",
   "gpu": "L4",
   "batch": 2,
   "params": 1080932352,
   "variantOf": "deepseek-v3",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 24.0652,
   "shippedGb": 24.1223,
   "accelerateFp32Gb": 17.2614,
   "accelerateBf16Gb": 8.6307
  },
  {
   "template": "deepseek-v3-r",
   "family": "transformer",
   "gpu": "L4",
   "batch": 4,
   "params": 1080932352,
   "variantOf": "deepseek-v3",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 27.47,
   "shippedGb": 27.5309,
   "accelerateFp32Gb": 17.2614,
   "accelerateBf16Gb": 8.6307
  },
  {
   "template": "deepseek-v3-r",
   "family": "transformer",
   "gpu": "L4",
   "batch": 8,
   "params": 1080932352,
   "variantOf": "deepseek-v3",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 34.2794,
   "shippedGb": 34.3481,
   "accelerateFp32Gb": 17.2614,
   "accelerateBf16Gb": 8.6307
  },
  {
   "template": "deepseek-v3-s",
   "family": "transformer",
   "gpu": "L4",
   "batch": 4,
   "params": 377169152,
   "variantOf": "deepseek-v3",
   "measuredPeakGb": 10.8446,
   "heldOutGb": 10.6233,
   "shippedGb": 10.6729,
   "accelerateFp32Gb": 6.0012,
   "accelerateBf16Gb": 3.0006
  },
  {
   "template": "deepseek-v3-s",
   "family": "transformer",
   "gpu": "L4",
   "batch": 8,
   "params": 377169152,
   "variantOf": "deepseek-v3",
   "measuredPeakGb": 16.4458,
   "heldOutGb": 14.0701,
   "shippedGb": 14.1444,
   "accelerateFp32Gb": 6.0012,
   "accelerateBf16Gb": 3.0006
  },
  {
   "template": "deepseek-v3-s",
   "family": "transformer",
   "gpu": "L4",
   "batch": 12,
   "params": 377169152,
   "variantOf": "deepseek-v3",
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 17.5168,
   "shippedGb": 17.6159,
   "accelerateFp32Gb": 6.0012,
   "accelerateBf16Gb": 3.0006
  },
  {
   "template": "deepseek-v3-s",
   "family": "transformer",
   "gpu": "L4",
   "batch": 16,
   "params": 377169152,
   "variantOf": "deepseek-v3",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 20.9636,
   "shippedGb": 21.0874,
   "accelerateFp32Gb": 6.0012,
   "accelerateBf16Gb": 3.0006
  },
  {
   "template": "diffusion-unet",
   "family": "conv",
   "gpu": "L4",
   "batch": 12,
   "params": 6687044,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 18.4172,
   "shippedGb": 19.6907,
   "accelerateFp32Gb": 0.107,
   "accelerateBf16Gb": 0.0535
  },
  {
   "template": "diffusion-unet",
   "family": "conv",
   "gpu": "L4",
   "batch": 16,
   "params": 6687044,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 24.5136,
   "shippedGb": 26.2115,
   "accelerateFp32Gb": 0.107,
   "accelerateBf16Gb": 0.0535
  },
  {
   "template": "diffusion-unet",
   "family": "conv",
   "gpu": "L4",
   "batch": 24,
   "params": 6687044,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 36.7064,
   "shippedGb": 39.253,
   "accelerateFp32Gb": 0.107,
   "accelerateBf16Gb": 0.0535
  },
  {
   "template": "dit-xl2",
   "family": "transformer",
   "gpu": "L4",
   "batch": 16,
   "params": 670684064,
   "measuredPeakGb": 16.6919,
   "heldOutGb": 21.9003,
   "shippedGb": 21.3397,
   "accelerateFp32Gb": 10.7309,
   "accelerateBf16Gb": 5.3655
  },
  {
   "template": "dit-xl2",
   "family": "transformer",
   "gpu": "L4",
   "batch": 20,
   "params": 670684064,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": 18.9505,
   "heldOutGb": 24.1249,
   "shippedGb": 23.4553,
   "accelerateFp32Gb": 10.7309,
   "accelerateBf16Gb": 5.3655
  },
  {
   "template": "dit-xl2",
   "family": "transformer",
   "gpu": "L4",
   "batch": 32,
   "params": 670684064,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 30.7987,
   "shippedGb": 29.8023,
   "accelerateFp32Gb": 10.7309,
   "accelerateBf16Gb": 5.3655
  },
  {
   "template": "dit-xl2",
   "family": "transformer",
   "gpu": "L4",
   "batch": 48,
   "params": 670684064,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 39.6971,
   "shippedGb": 38.2648,
   "accelerateFp32Gb": 10.7309,
   "accelerateBf16Gb": 5.3655
  },
  {
   "template": "eeg-conformer",
   "family": "vision-transformer",
   "gpu": "L4",
   "batch": 3072,
   "params": 87204,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": 15.726,
   "heldOutGb": 20.0269,
   "shippedGb": 18.2242,
   "accelerateFp32Gb": 0.0013,
   "accelerateBf16Gb": 0.0006
  },
  {
   "template": "eeg-conformer",
   "family": "vision-transformer",
   "gpu": "L4",
   "batch": 4096,
   "params": 87204,
   "measuredPeakGb": 19.4586,
   "heldOutGb": 26.7021,
   "shippedGb": 24.2984,
   "accelerateFp32Gb": 0.0013,
   "accelerateBf16Gb": 0.0006
  },
  {
   "template": "gpt2",
   "family": "embedding-mlp",
   "gpu": "L4",
   "batch": 40,
   "params": 84334417,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 18.2437,
   "shippedGb": 18.5718,
   "accelerateFp32Gb": 1.3493,
   "accelerateBf16Gb": 0.6747
  },
  {
   "template": "gpt2",
   "family": "embedding-mlp",
   "gpu": "L4",
   "batch": 48,
   "params": 84334417,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 21.5696,
   "shippedGb": 21.9624,
   "accelerateFp32Gb": 1.3493,
   "accelerateBf16Gb": 0.6747
  },
  {
   "template": "gpt2",
   "family": "embedding-mlp",
   "gpu": "L4",
   "batch": 64,
   "params": 84334417,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 28.2214,
   "shippedGb": 28.7434,
   "accelerateFp32Gb": 1.3493,
   "accelerateBf16Gb": 0.6747
  },
  {
   "template": "gpt2",
   "family": "embedding-mlp",
   "gpu": "L4",
   "batch": 96,
   "params": 84334417,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 41.525,
   "shippedGb": 42.3056,
   "accelerateFp32Gb": 1.3493,
   "accelerateBf16Gb": 0.6747
  },
  {
   "template": "hstu",
   "family": "other",
   "gpu": "L4",
   "batch": 12,
   "params": 210057728,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 20.0855,
   "shippedGb": 20.1002,
   "accelerateFp32Gb": 3.3608,
   "accelerateBf16Gb": 1.6804
  },
  {
   "template": "hstu",
   "family": "other",
   "gpu": "L4",
   "batch": 16,
   "params": 210057728,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 25.4398,
   "shippedGb": 25.456,
   "accelerateFp32Gb": 3.3608,
   "accelerateBf16Gb": 1.6804
  },
  {
   "template": "hstu",
   "family": "other",
   "gpu": "L4",
   "batch": 24,
   "params": 210057728,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 36.1484,
   "shippedGb": 36.1675,
   "accelerateFp32Gb": 3.3608,
   "accelerateBf16Gb": 1.6804
  },
  {
   "template": "ijepa",
   "family": "other",
   "gpu": "L4",
   "batch": 640,
   "params": 40770432,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 18.3596,
   "shippedGb": 18.5534,
   "accelerateFp32Gb": 0.6518,
   "accelerateBf16Gb": 0.3259
  },
  {
   "template": "ijepa",
   "family": "other",
   "gpu": "L4",
   "batch": 768,
   "params": 40770432,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 21.875,
   "shippedGb": 22.1076,
   "accelerateFp32Gb": 0.6518,
   "accelerateBf16Gb": 0.3259
  },
  {
   "template": "ijepa",
   "family": "other",
   "gpu": "L4",
   "batch": 4096,
   "params": 40770432,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 113.2762,
   "shippedGb": 114.5175,
   "accelerateFp32Gb": 0.6518,
   "accelerateBf16Gb": 0.3259
  },
  {
   "template": "llama3-block",
   "family": "transformer",
   "gpu": "L4",
   "batch": 4,
   "params": 702554112,
   "measuredPeakGb": 14.2024,
   "heldOutGb": 14.7629,
   "shippedGb": 14.808,
   "accelerateFp32Gb": 11.2409,
   "accelerateBf16Gb": 5.6204
  },
  {
   "template": "llama3-block",
   "family": "transformer",
   "gpu": "L4",
   "batch": 16,
   "params": 702554112,
   "measuredPeakGb": 18.9403,
   "heldOutGb": 18.6662,
   "shippedGb": 18.7648,
   "accelerateFp32Gb": 11.2409,
   "accelerateBf16Gb": 5.6204
  },
  {
   "template": "llama3-block",
   "family": "transformer",
   "gpu": "L4",
   "batch": 20,
   "params": 702554112,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 19.9673,
   "shippedGb": 20.0837,
   "accelerateFp32Gb": 11.2409,
   "accelerateBf16Gb": 5.6204
  },
  {
   "template": "llama3-block",
   "family": "transformer",
   "gpu": "L4",
   "batch": 24,
   "params": 702554112,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 21.2684,
   "shippedGb": 21.4027,
   "accelerateFp32Gb": 11.2409,
   "accelerateBf16Gb": 5.6204
  },
  {
   "template": "matryoshka-embed",
   "family": "other",
   "gpu": "L4",
   "batch": 384,
   "params": 54745345,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 19.1654,
   "shippedGb": 21.0922,
   "accelerateFp32Gb": 0.8759,
   "accelerateBf16Gb": 0.4379
  },
  {
   "template": "matryoshka-embed",
   "family": "other",
   "gpu": "L4",
   "batch": 512,
   "params": 54745345,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 25.2018,
   "shippedGb": 27.7726,
   "accelerateFp32Gb": 0.8759,
   "accelerateBf16Gb": 0.4379
  },
  {
   "template": "matryoshka-embed",
   "family": "other",
   "gpu": "L4",
   "batch": 768,
   "params": 54745345,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 37.2747,
   "shippedGb": 41.1334,
   "accelerateFp32Gb": 0.8759,
   "accelerateBf16Gb": 0.4379
  },
  {
   "template": "mixtral-block",
   "family": "transformer",
   "gpu": "L4",
   "batch": 1,
   "params": 981508096,
   "excluded": "measured on an export whose experts had two matrices where the graph has three (0.98B parameters built against 1.45B planned); the exporter now builds three, and the corrected model is past a 24 GB card on weights alone",
   "measuredPeakGb": 19.7814,
   "heldOutGb": 29.0013,
   "shippedGb": 29.0761,
   "accelerateFp32Gb": 23.2203,
   "accelerateBf16Gb": 11.6102
  },
  {
   "template": "patch-tst",
   "family": "transformer",
   "gpu": "L4",
   "batch": 12288,
   "params": 329860,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 19.6598,
   "shippedGb": 19.9757,
   "accelerateFp32Gb": 0.0063,
   "accelerateBf16Gb": 0.0032
  },
  {
   "template": "qwen2-vl",
   "family": "other",
   "gpu": "L4",
   "batch": 1,
   "params": 1549126912,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 30.1918,
   "shippedGb": 30.2288,
   "accelerateFp32Gb": 24.7862,
   "accelerateBf16Gb": 12.3931
  },
  {
   "template": "qwen3-8b-r",
   "family": "transformer",
   "gpu": "L4",
   "batch": 1,
   "params": 648072448,
   "variantOf": "qwen3-8b",
   "measuredPeakGb": 13.2407,
   "heldOutGb": 14.4778,
   "shippedGb": 14.471,
   "accelerateFp32Gb": 10.3692,
   "accelerateBf16Gb": 5.1846
  },
  {
   "template": "qwen3-8b-r",
   "family": "transformer",
   "gpu": "L4",
   "batch": 2,
   "params": 648072448,
   "variantOf": "qwen3-8b",
   "measuredPeakGb": 15.1551,
   "heldOutGb": 16.5031,
   "shippedGb": 16.499,
   "accelerateFp32Gb": 10.3692,
   "accelerateBf16Gb": 5.1846
  },
  {
   "template": "qwen3-8b-r",
   "family": "transformer",
   "gpu": "L4",
   "batch": 3,
   "params": 648072448,
   "variantOf": "qwen3-8b",
   "measuredPeakGb": 18.3197,
   "heldOutGb": 18.5284,
   "shippedGb": 18.527,
   "accelerateFp32Gb": 10.3692,
   "accelerateBf16Gb": 5.1846
  },
  {
   "template": "qwen3-8b-r",
   "family": "transformer",
   "gpu": "L4",
   "batch": 4,
   "params": 648072448,
   "variantOf": "qwen3-8b",
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 20.5537,
   "shippedGb": 20.555,
   "accelerateFp32Gb": 10.3692,
   "accelerateBf16Gb": 5.1846
  },
  {
   "template": "simple-rnn",
   "family": "recurrent",
   "gpu": "L4",
   "batch": 6144,
   "params": 1100298,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": 10.6332,
   "heldOutGb": 41.1887,
   "shippedGb": 25.1256,
   "accelerateFp32Gb": 0.0176,
   "accelerateBf16Gb": 0.0088
  },
  {
   "template": "swin-tiny",
   "family": "transformer",
   "gpu": "L4",
   "batch": 24,
   "params": 28260808,
   "measuredPeakGb": 16.43,
   "heldOutGb": 17.1614,
   "shippedGb": 16.842,
   "accelerateFp32Gb": 0.4522,
   "accelerateBf16Gb": 0.2261
  },
  {
   "template": "swin-tiny",
   "family": "transformer",
   "gpu": "L4",
   "batch": 48,
   "params": 28260808,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 33.7818,
   "shippedGb": 33.1413,
   "accelerateFp32Gb": 0.4522,
   "accelerateBf16Gb": 0.2261
  },
  {
   "template": "swin-tiny",
   "family": "transformer",
   "gpu": "L4",
   "batch": 64,
   "params": 28260808,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 44.862,
   "shippedGb": 44.0075,
   "accelerateFp32Gb": 0.4522,
   "accelerateBf16Gb": 0.2261
  },
  {
   "template": "t5-small",
   "family": "transformer",
   "gpu": "L4",
   "batch": 320,
   "params": 56736640,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": 16.7659,
   "heldOutGb": 19.2974,
   "shippedGb": 18.7959,
   "accelerateFp32Gb": 0.9078,
   "accelerateBf16Gb": 0.4539
  },
  {
   "template": "t5-small",
   "family": "transformer",
   "gpu": "L4",
   "batch": 384,
   "params": 56736640,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 22.936,
   "shippedGb": 22.3372,
   "accelerateFp32Gb": 0.9078,
   "accelerateBf16Gb": 0.4539
  },
  {
   "template": "t5-small",
   "family": "transformer",
   "gpu": "L4",
   "batch": 640,
   "params": 56736640,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 37.4904,
   "shippedGb": 36.5024,
   "accelerateFp32Gb": 0.9078,
   "accelerateBf16Gb": 0.4539
  },
  {
   "template": "transformer-block",
   "family": "transformer",
   "gpu": "L4",
   "batch": 512,
   "params": 7087872,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 19.6121,
   "shippedGb": 19.8187,
   "accelerateFp32Gb": 0.1134,
   "accelerateBf16Gb": 0.0567
  },
  {
   "template": "transformer-block",
   "family": "transformer",
   "gpu": "L4",
   "batch": 640,
   "params": 7087872,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 24.4812,
   "shippedGb": 24.7394,
   "accelerateFp32Gb": 0.1134,
   "accelerateBf16Gb": 0.0567
  },
  {
   "template": "transformer-block",
   "family": "transformer",
   "gpu": "L4",
   "batch": 768,
   "params": 7087872,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 29.3502,
   "shippedGb": 29.66,
   "accelerateFp32Gb": 0.1134,
   "accelerateBf16Gb": 0.0567
  },
  {
   "template": "transformer-block",
   "family": "transformer",
   "gpu": "L4",
   "batch": 1024,
   "params": 7087872,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 39.0883,
   "shippedGb": 39.5014,
   "accelerateFp32Gb": 0.1134,
   "accelerateBf16Gb": 0.0567
  },
  {
   "template": "unet",
   "family": "conv",
   "gpu": "L4",
   "batch": 48,
   "params": 720705,
   "measuredPeakGb": 10.4225,
   "heldOutGb": 15.7924,
   "shippedGb": 11.0189,
   "accelerateFp32Gb": 0.0115,
   "accelerateBf16Gb": 0.0058
  },
  {
   "template": "unet",
   "family": "conv",
   "gpu": "L4",
   "batch": 64,
   "params": 720705,
   "measuredPeakGb": 13.8933,
   "heldOutGb": 21.0519,
   "shippedGb": 14.6872,
   "accelerateFp32Gb": 0.0115,
   "accelerateBf16Gb": 0.0058
  },
  {
   "template": "unet",
   "family": "conv",
   "gpu": "L4",
   "batch": 96,
   "params": 720705,
   "measuredPeakGb": 20.8348,
   "heldOutGb": 31.571,
   "shippedGb": 22.0239,
   "accelerateFp32Gb": 0.0115,
   "accelerateBf16Gb": 0.0058
  },
  {
   "template": "video-dit",
   "family": "transformer",
   "gpu": "L4",
   "batch": 24,
   "params": 99434192,
   "measuredPeakGb": 12.2911,
   "heldOutGb": 12.5139,
   "shippedGb": 12.4457,
   "accelerateFp32Gb": 1.5889,
   "accelerateBf16Gb": 0.7944
  },
  {
   "template": "video-dit",
   "family": "transformer",
   "gpu": "L4",
   "batch": 32,
   "params": 99434192,
   "measuredPeakGb": 15.9165,
   "heldOutGb": 16.0515,
   "shippedGb": 15.9587,
   "accelerateFp32Gb": 1.5889,
   "accelerateBf16Gb": 0.7944
  },
  {
   "template": "video-dit",
   "family": "transformer",
   "gpu": "L4",
   "batch": 40,
   "params": 99434192,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": 19.5404,
   "heldOutGb": 19.5891,
   "shippedGb": 19.4717,
   "accelerateFp32Gb": 1.5889,
   "accelerateBf16Gb": 0.7944
  },
  {
   "template": "video-dit",
   "family": "transformer",
   "gpu": "L4",
   "batch": 48,
   "params": 99434192,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 23.1268,
   "shippedGb": 22.9847,
   "accelerateFp32Gb": 1.5889,
   "accelerateBf16Gb": 0.7944
  },
  {
   "template": "vit-b16",
   "family": "transformer",
   "gpu": "L4",
   "batch": 1536,
   "params": 8449000,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 18.8597,
   "shippedGb": 18.9817,
   "accelerateFp32Gb": 0.1352,
   "accelerateBf16Gb": 0.0676
  },
  {
   "template": "vit-b16",
   "family": "transformer",
   "gpu": "L4",
   "batch": 2048,
   "params": 8449000,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 25.0924,
   "shippedGb": 25.2548,
   "accelerateFp32Gb": 0.1352,
   "accelerateBf16Gb": 0.0676
  },
  {
   "template": "vit-b16",
   "family": "transformer",
   "gpu": "L4",
   "batch": 4096,
   "params": 8449000,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 50.0228,
   "shippedGb": 50.3474,
   "accelerateFp32Gb": 0.1352,
   "accelerateBf16Gb": 0.0676
  },
  {
   "template": "whisper-small",
   "family": "embedding-mlp",
   "gpu": "L4",
   "batch": 96,
   "params": 48703129,
   "measuredAfterFit": "refit check, 2026-10-17",
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 19.2943,
   "shippedGb": 19.3181,
   "accelerateFp32Gb": 0.7518,
   "accelerateBf16Gb": 0.3759
  },
  {
   "template": "whisper-small",
   "family": "embedding-mlp",
   "gpu": "L4",
   "batch": 192,
   "params": 48703129,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 37.6888,
   "shippedGb": 37.7341,
   "accelerateFp32Gb": 0.7518,
   "accelerateBf16Gb": 0.3759
  },
  {
   "template": "whisper-small",
   "family": "embedding-mlp",
   "gpu": "L4",
   "batch": 320,
   "params": 48703129,
   "measuredPeakGb": null,
   "oomOnGb": 24,
   "heldOutGb": 62.2148,
   "shippedGb": 62.2887,
   "accelerateFp32Gb": 0.7518,
   "accelerateBf16Gb": 0.3759
  }
 ]
}
