ai.onnx.Mul / build /webgpu /bench.json
Xenova's picture
Xenova HF Staff
sync 91d990483a17
81259c2 verified
Raw
History Blame
2.27 kB
{
"tunableSpace": { "WORKGROUP_SIZE": [64, 128, 256] },
"cases": [
{
"name": "mul-f32-4m",
"preset": "smoke",
"vars": { "dtype": "float32", "count": 4194304 },
"inputs": {
"a": { "shape": [4194304], "dtype": "float32", "dist": "normal", "seed": 202, "scale": 2 },
"b": { "shape": [4194304], "dtype": "float32", "dist": "normal", "seed": 203, "scale": 2 }
},
"outputs": { "c": { "shape": [4194304], "dtype": "float32" } },
"bench": {
"primary": true,
"metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 3" }]
}
},
{
"name": "mul-f32-broadcast-vec4-4m",
"preset": "smoke",
"vars": { "dtype": "float32", "count": 4194304 },
"inputs": {
"a": { "shape": [4096, 1024], "dtype": "float32", "dist": "normal", "seed": 210, "scale": 2 },
"b": { "shape": [1, 1024], "dtype": "float32", "dist": "normal", "seed": 211, "scale": 2 }
},
"outputs": { "c": { "shape": [4096, 1024], "dtype": "float32" } },
"bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] }
},
{
"name": "mul-f32-scalar-broadcast-fallback-4m",
"preset": "smoke",
"vars": { "dtype": "float32", "count": 4190208 },
"inputs": {
"a": { "shape": [4096, 1023], "dtype": "float32", "dist": "normal", "seed": 212, "scale": 2 },
"b": { "shape": [1, 1023], "dtype": "float32", "dist": "normal", "seed": 213, "scale": 2 }
},
"outputs": { "c": { "shape": [4096, 1023], "dtype": "float32" } },
"bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 2" }] }
},
{
"name": "mul-f16-4m",
"preset": "smoke",
"vars": { "dtype": "float16", "count": 4194304 },
"inputs": {
"a": { "shape": [4194304], "dtype": "float16", "dist": "normal", "seed": 214, "scale": 2 },
"b": { "shape": [4194304], "dtype": "float16", "dist": "normal", "seed": 215, "scale": 2 }
},
"outputs": { "c": { "shape": [4194304], "dtype": "float16" } },
"bench": { "metrics": [{ "type": "bandwidth", "value": "args.count * dtypeBytes(args.dtype) * 3" }] }
}
]
}