ai.onnx.ArgMin / build /webgpu /bench.json
Xenova's picture
Xenova HF Staff
sync 2e7068faf55e
7c6615c verified
Raw
History Blame
4.82 kB
{
"op": "ai.onnx.ArgMin",
"cases": [
{
"name": "argmin-axis0-f32-1024x512",
"preset": "smoke",
"vars": { "dtype": "float32", "rows": 1024, "cols": 512 },
"attrs": { "axis": 0, "keepdims": 0 },
"inputs": { "x": { "shape": [1024, 512], "dtype": "float32", "dist": "normal", "seed": 977, "scale": 1 } },
"outputs": { "y": { "shape": [512], "dtype": "uint32" } },
"bench": {
"primary": true,
"metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * dtypeBytes(args.dtype)" }]
}
},
{
"name": "scores",
"attrs": { "axis": 1, "keepdims": 0 },
"inputs": { "x": { "dtype": "float32", "shape": [1024, 768] } },
"outputs": { "y": { "dtype": "uint32", "shape": [1024] } }
},
{
"name": "argmin-axis0-large-65536x512",
"preset": "smoke",
"vars": { "dtype": "float32", "rows": 65536, "cols": 512 },
"attrs": { "axis": 0, "keepdims": 0 },
"inputs": { "x": { "shape": [65536, 512], "dtype": "float32", "dist": "normal", "seed": 211, "scale": 0.2 } },
"outputs": { "y": { "shape": [512], "dtype": "uint32" } },
"bench": { "primary": true, "metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * 4" }] }
},
{
"name": "argmin-lastaxis-f32-4096x4096",
"preset": "smoke",
"vars": { "rows": 4096, "cols": 4096 },
"attrs": { "axis": 1, "keepdims": 0 },
"inputs": { "x": { "shape": [4096, 4096], "dtype": "float32", "dist": "normal", "seed": 117, "scale": 0.2 } },
"outputs": { "y": { "shape": [4096], "dtype": "uint32" } },
"bench": { "primary": true, "metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * 4" }] }
},
{
"name": "argmin-lastaxis-f32-4096x4095",
"preset": "smoke",
"vars": { "dtype": "float32", "rows": 4096, "cols": 4095 },
"attrs": { "axis": 1, "keepdims": 0 },
"inputs": { "x": { "shape": [4096, 4095], "dtype": "float32", "dist": "normal", "seed": 419, "scale": 0.2 } },
"outputs": { "y": { "shape": [4096], "dtype": "uint32" } },
"bench": { "metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * dtypeBytes(args.dtype)" }] }
},
{
"name": "argmin-vocab-unaligned-f32-1x50257",
"preset": "smoke",
"vars": { "dtype": "float32", "rows": 1, "cols": 50257 },
"attrs": { "axis": 1, "keepdims": 0 },
"inputs": { "x": { "shape": [1, 50257], "dtype": "float32", "dist": "normal", "seed": 743, "scale": 1 } },
"outputs": { "y": { "shape": [1], "dtype": "uint32" } },
"bench": { "metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * dtypeBytes(args.dtype)" }] }
},
{
"name": "argmin-lastaxis-f16-1024x768",
"preset": "smoke",
"vars": { "dtype": "float16", "rows": 1024, "cols": 768 },
"attrs": { "axis": 1, "keepdims": 0 },
"inputs": { "x": { "shape": [1024, 768], "dtype": "float16", "dist": "normal", "seed": 523, "scale": 0.2 } },
"outputs": { "y": { "shape": [1024], "dtype": "uint32" } },
"bench": { "metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * dtypeBytes(args.dtype)" }] }
},
{
"name": "argmin-axis0-f32-32x512",
"preset": "smoke",
"vars": { "dtype": "float32", "rows": 32, "cols": 512 },
"attrs": { "axis": 0, "keepdims": 0 },
"inputs": { "x": { "shape": [32, 512], "dtype": "float32", "dist": "normal", "seed": 631, "scale": 0.2 } },
"outputs": { "y": { "shape": [512], "dtype": "uint32" } },
"bench": { "metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * dtypeBytes(args.dtype)" }] }
},
{
"name": "argmin-lastaxis-uint8-2048x4096-serial",
"preset": "stress",
"vars": { "dtype": "uint8", "rows": 2048, "cols": 4096 },
"attrs": { "axis": 1, "keepdims": 0 },
"inputs": {
"x": { "shape": [2048, 4096], "dtype": "uint8", "dist": "uniform", "seed": 743, "min": 0, "max": 255 }
},
"outputs": { "y": { "shape": [2048], "dtype": "uint32" } },
"bench": { "metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * dtypeBytes(args.dtype)" }] }
},
{
"name": "argmin-axis0-narrow-starvation-131072x8",
"preset": "smoke",
"vars": { "dtype": "float32", "rows": 131072, "cols": 8 },
"attrs": { "axis": 0, "keepdims": 0 },
"inputs": { "x": { "shape": [131072, 8], "dtype": "float32", "dist": "normal", "seed": 827, "scale": 1 } },
"outputs": { "y": { "shape": [8], "dtype": "uint32" } },
"bench": {
"primary": true,
"metrics": [{ "type": "bandwidth", "value": "args.rows * args.cols * dtypeBytes(args.dtype)" }]
}
}
],
"tunableSpace": { "WORKGROUP_SIZE": [64, 128, 256] }
}