ai.onnx.Split / build /webgpu /bench.json
Xenova's picture
Xenova HF Staff
sync 2e7068faf55e
854303d verified
Raw
History Blame
5.56 kB
{
"op": "ai.onnx.Split",
"cases": [
{
"name": "qkv_like",
"attrs": { "axis": 2 },
"inputs": { "input": { "dtype": "float32", "shape": [1, 128, 2304] } },
"outputs": {
"y0": { "dtype": "float32", "shape": [1, 128, 768] },
"y1": { "dtype": "float32", "shape": [1, 128, 1536] }
},
"preset": "smoke",
"vars": { "count": 294912, "dtype": "float32" },
"bench": {
"primary": true,
"metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }]
}
},
{
"name": "three-way-equal-axis1-m1500-n1152",
"preset": "model",
"attrs": { "axis": 1 },
"vars": { "count": 1728000, "dtype": "float32" },
"inputs": { "input": { "shape": [1500, 1152], "dtype": "float32", "dist": "normal", "seed": 976, "scale": 0.5 } },
"outputs": {
"y0": { "shape": [1500, 384], "dtype": "float32" },
"y1": { "shape": [1500, 384], "dtype": "float32" },
"y2": { "shape": [1500, 384], "dtype": "float32" }
},
"bench": {
"primary": true,
"metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }]
}
},
{
"name": "split-block-axis0-f32-8m",
"preset": "smoke",
"attrs": { "axis": 0 },
"vars": { "count": 8388608, "dtype": "float32" },
"inputs": { "input": { "shape": [8192, 1024], "dtype": "float32", "dist": "normal", "seed": 970, "scale": 1 } },
"outputs": {
"y0": { "shape": [4096, 1024], "dtype": "float32" },
"y1": { "shape": [4096, 1024], "dtype": "float32" }
},
"bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] }
},
{
"name": "split-scalar-cliff-last-axis-unaligned",
"attrs": { "axis": 2 },
"inputs": { "input": { "dtype": "float32", "shape": [1, 128, 2306] } },
"outputs": {
"y0": { "dtype": "float32", "shape": [1, 128, 770] },
"y1": { "dtype": "float32", "shape": [1, 128, 1536] }
},
"preset": "smoke",
"vars": { "count": 295168, "dtype": "float32" },
"bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] }
},
{
"name": "qkv_like_f16",
"attrs": { "axis": 2 },
"inputs": { "input": { "dtype": "float16", "shape": [1, 128, 2304] } },
"outputs": {
"y0": { "dtype": "float16", "shape": [1, 128, 768] },
"y1": { "dtype": "float16", "shape": [1, 128, 1536] }
},
"preset": "smoke",
"vars": { "count": 294912, "dtype": "float16" },
"bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] }
},
{
"name": "split-five-outputs-scalar-8m",
"attrs": { "axis": 1 },
"inputs": { "input": { "dtype": "float32", "shape": [8, 1310720], "dist": "normal", "seed": 971, "scale": 1 } },
"outputs": {
"y0": { "dtype": "float32", "shape": [8, 262144] },
"y1": { "dtype": "float32", "shape": [8, 262144] },
"y2": { "dtype": "float32", "shape": [8, 262144] },
"y3": { "dtype": "float32", "shape": [8, 262144] },
"y4": { "dtype": "float32", "shape": [8, 262144] }
},
"preset": "smoke",
"vars": { "count": 10485760, "dtype": "float32" },
"bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] }
},
{
"name": "split-scalar-rank7-middle-axis-2m",
"preset": "stress",
"attrs": { "axis": 2 },
"vars": { "count": 1990656, "dtype": "float32" },
"inputs": {
"input": { "dtype": "float32", "shape": [64, 64, 6, 3, 3, 3, 3], "dist": "normal", "seed": 973, "scale": 1 }
},
"outputs": {
"y0": { "dtype": "float32", "shape": [64, 64, 3, 3, 3, 3, 3] },
"y1": { "dtype": "float32", "shape": [64, 64, 3, 3, 3, 3, 3] }
},
"bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] }
},
{
"name": "split-scalar-four-outputs-unaligned-runs-2m",
"preset": "stress",
"attrs": { "axis": 1 },
"vars": { "count": 2211840, "dtype": "float32" },
"inputs": {
"input": { "dtype": "float32", "shape": [8192, 10, 3, 3, 3], "dist": "normal", "seed": 974, "scale": 1 }
},
"outputs": {
"y0": { "dtype": "float32", "shape": [8192, 3, 3, 3, 3] },
"y1": { "dtype": "float32", "shape": [8192, 3, 3, 3, 3] },
"y2": { "dtype": "float32", "shape": [8192, 2, 3, 3, 3] },
"y3": { "dtype": "float32", "shape": [8192, 2, 3, 3, 3] }
},
"bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] }
},
{
"name": "split-three-output-q2050-kv512-odd-segment-pathology",
"preset": "stress",
"attrs": { "axis": 2 },
"vars": { "count": 1573888, "dtype": "float32" },
"inputs": { "input": { "dtype": "float32", "shape": [1, 512, 3074], "dist": "normal", "seed": 975, "scale": 1 } },
"outputs": {
"y0": { "dtype": "float32", "shape": [1, 512, 2050] },
"y1": { "dtype": "float32", "shape": [1, 512, 512] },
"y2": { "dtype": "float32", "shape": [1, 512, 512] }
},
"bench": {
"primary": true,
"metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }]
}
}
],
"tunableSpace": { "WORKGROUP_SIZE": [64, 128, 256] }
}