{ "op": "ai.onnx.RMSNormalization", "cases": [ { "name": "rmsnorm-f32-256x1024", "preset": "smoke", "vars": { "dtype": "float32", "rows": 256, "dim": 1024 }, "attrs": { "epsilon": 0.000001, "axis": -1 }, "inputs": { "x": { "shape": [256, 1024], "dtype": "float32", "dist": "normal", "seed": 710, "scale": 0.5 }, "scale": { "shape": [1024], "dtype": "float32", "dist": "uniform", "seed": 711, "scale": 0.25, "offset": 1 } }, "outputs": { "y": { "shape": [256, 1024], "dtype": "float32" } }, "bench": { "primary": true, "metrics": [{ "type": "bandwidth", "value": "(args.rows * args.dim * 2 + args.dim) * dtypeBytes(args.dtype)" }] } }, { "name": "rmsnorm-f32-4096x4096", "preset": "smoke", "vars": { "dtype": "float32", "rows": 4096, "dim": 4096 }, "attrs": { "epsilon": 0.000001, "axis": -1 }, "inputs": { "x": { "shape": [4096, 4096], "dtype": "float32", "dist": "normal", "seed": 712, "scale": 0.5 }, "scale": { "shape": [4096], "dtype": "float32", "dist": "uniform", "seed": 713, "scale": 0.25, "offset": 1 } }, "outputs": { "y": { "shape": [4096, 4096], "dtype": "float32" } }, "bench": { "primary": true, "metrics": [{ "type": "bandwidth", "value": "(args.rows * args.dim * 2 + args.dim) * dtypeBytes(args.dtype)" }] } }, { "name": "rmsnorm-f16-4096x4096", "preset": "smoke", "vars": { "dtype": "float16", "rows": 4096, "dim": 4096 }, "attrs": { "epsilon": 0.000001, "axis": -1 }, "inputs": { "x": { "shape": [4096, 4096], "dtype": "float16", "dist": "normal", "seed": 714, "scale": 0.5 }, "scale": { "shape": [4096], "dtype": "float16", "dist": "uniform", "seed": 715, "scale": 0.25, "offset": 1 } }, "outputs": { "y": { "shape": [4096, 4096], "dtype": "float16" } }, "bench": { "metrics": [{ "type": "bandwidth", "value": "(args.rows * args.dim * 2 + args.dim) * dtypeBytes(args.dtype)" }] } }, { "name": "rmsnorm-f16-1x4096-rows1-decode", "preset": "smoke", "vars": { "dtype": "float16", "rows": 1, "dim": 4096 }, "attrs": { "epsilon": 0.000001, "axis": -1 }, "inputs": { "x": { "shape": [1, 4096], "dtype": "float16", "dist": "normal", "seed": 7200, "scale": 0.5 }, "scale": { "shape": [4096], "dtype": "float16", "dist": "uniform", "seed": 7201, "scale": 0.25, "offset": 1 } }, "outputs": { "y": { "shape": [1, 4096], "dtype": "float16" } }, "bench": { "metrics": [{ "type": "bandwidth", "value": "(args.rows * args.dim * 2 + args.dim) * dtypeBytes(args.dtype)" }] } }, { "name": "rmsnorm-f16-256x4096-aligned-vec4", "preset": "smoke", "vars": { "dtype": "float16", "rows": 256, "dim": 4096 }, "attrs": { "epsilon": 0.000001, "axis": -1 }, "inputs": { "x": { "shape": [256, 4096], "dtype": "float16", "dist": "normal", "seed": 7210, "scale": 0.5 }, "scale": { "shape": [4096], "dtype": "float16", "dist": "uniform", "seed": 7211, "scale": 0.25, "offset": 1 } }, "outputs": { "y": { "shape": [256, 4096], "dtype": "float16" } }, "bench": { "metrics": [{ "type": "bandwidth", "value": "(args.rows * args.dim * 2 + args.dim) * dtypeBytes(args.dtype)" }] } }, { "name": "rmsnorm-f16-256x4094-unaligned-alignmentcliff", "preset": "smoke", "vars": { "dtype": "float16", "rows": 256, "dim": 4094 }, "attrs": { "epsilon": 0.000001, "axis": -1 }, "inputs": { "x": { "shape": [256, 4094], "dtype": "float16", "dist": "normal", "seed": 7220, "scale": 0.5 }, "scale": { "shape": [4094], "dtype": "float16", "dist": "uniform", "seed": 7221, "scale": 0.25, "offset": 1 } }, "outputs": { "y": { "shape": [256, 4094], "dtype": "float16" } }, "bench": { "metrics": [{ "type": "bandwidth", "value": "(args.rows * args.dim * 2 + args.dim) * dtypeBytes(args.dtype)" }] } }, { "name": "rmsnorm-f32-8x256x1024-suffixaxis1-generic", "preset": "smoke", "vars": { "dtype": "float32", "outer": 8, "normSize": 262144 }, "attrs": { "epsilon": 0.000001, "axis": 1 }, "inputs": { "x": { "shape": [8, 256, 1024], "dtype": "float32", "dist": "normal", "seed": 7230, "scale": 0.5 }, "scale": { "shape": [256, 1024], "dtype": "float32", "dist": "uniform", "seed": 7231, "scale": 0.25, "offset": 1 } }, "outputs": { "y": { "shape": [8, 256, 1024], "dtype": "float32" } }, "bench": { "metrics": [ { "type": "bandwidth", "value": "(args.outer * args.normSize * 2 + args.normSize) * dtypeBytes(args.dtype)" } ] } }, { "name": "rmsnorm-f32-65535x512-undercliff", "preset": "stress", "vars": { "dtype": "float32", "rows": 65535, "dim": 512 }, "attrs": { "epsilon": 0.000001, "axis": -1 }, "inputs": { "x": { "shape": [65535, 512], "dtype": "float32", "dist": "normal", "seed": 7240, "scale": 0.5 }, "scale": { "shape": [512], "dtype": "float32", "dist": "uniform", "seed": 7241, "scale": 0.25, "offset": 1 } }, "outputs": { "y": { "shape": [65535, 512], "dtype": "float32" } }, "bench": { "metrics": [{ "type": "bandwidth", "value": "(args.rows * args.dim * 2 + args.dim) * dtypeBytes(args.dtype)" }] } }, { "name": "rmsnorm-f32-70000x512-rows-dispatchcliff", "preset": "stress", "provenance": { "notes": "Stress-only dispatch/capacity case: declared tensors occupy 273 MiB of GPU storage." }, "vars": { "dtype": "float32", "rows": 70000, "dim": 512 }, "attrs": { "epsilon": 0.000001, "axis": -1 }, "inputs": { "x": { "shape": [70000, 512], "dtype": "float32", "dist": "normal", "seed": 7250, "scale": 0.5 }, "scale": { "shape": [512], "dtype": "float32", "dist": "uniform", "seed": 7251, "scale": 0.25, "offset": 1 } }, "outputs": { "y": { "shape": [70000, 512], "dtype": "float32" } }, "bench": { "metrics": [{ "type": "bandwidth", "value": "(args.rows * args.dim * 2 + args.dim) * dtypeBytes(args.dtype)" }] } }, { "name": "rmsnorm-f16-8x262144-suffixaxis0-nosplitk-occupancy-starve", "preset": "stress", "vars": { "dtype": "float16", "outer": 8, "normSize": 262144 }, "attrs": { "epsilon": 0.000001, "axis": 0 }, "inputs": { "x": { "shape": [8, 262144], "dtype": "float16", "dist": "normal", "seed": 7260, "scale": 0.5 }, "scale": { "shape": [262144], "dtype": "float16", "dist": "uniform", "seed": 7261, "scale": 0.25, "offset": 1 } }, "outputs": { "y": { "shape": [8, 262144], "dtype": "float16", "dist": "empty" } }, "bench": { "metrics": [ { "type": "bandwidth", "value": "(args.outer * args.normSize * 2 + args.normSize) * dtypeBytes(args.dtype)" } ] } }, { "name": "rmsnorm-f32-4x262144-lastaxis-scalarscale-fallback-starve", "preset": "stress", "vars": { "dtype": "float32", "rows": 4, "dim": 262144 }, "attrs": { "epsilon": 0.000001, "axis": -1 }, "inputs": { "x": { "shape": [4, 262144], "dtype": "float32", "dist": "normal", "seed": 7270, "scale": 0.5 }, "scale": { "shape": [1], "dtype": "float32", "dist": "uniform", "seed": 7271, "scale": 0.25, "offset": 1 } }, "outputs": { "y": { "shape": [4, 262144], "dtype": "float32", "dist": "empty" } }, "bench": { "metrics": [{ "type": "bandwidth", "value": "(args.rows * args.dim * 2 + args.dim) * dtypeBytes(args.dtype)" }] } } ] }