{ "op": "com.microsoft.BiasSplitGelu", "tunableSpace": { "WORKGROUP_SIZE": [64, 128, 256] }, "cases": [ { "name": "biassplitgelu-f32-4096x5120", "preset": "smoke", "vars": { "dtype": "float32" }, "inputs": { "X": { "shape": [1, 4096, 5120], "dtype": "float32", "dist": "normal", "seed": 610, "scale": 2 }, "bias": { "shape": [5120], "dtype": "float32", "dist": "normal", "seed": 611, "scale": 1 } }, "outputs": { "Y": { "shape": [1, 4096, 2560], "dtype": "float32" } }, "bench": { "metrics": [ { "type": "bandwidth", "value": "4096 * 5120 * dtypeBytes(args.dtype) + 4096 * 2560 * dtypeBytes(args.dtype)" } ] } }, { "name": "biassplitgelu-f16-4096x5120", "preset": "model", "vars": { "dtype": "float16" }, "inputs": { "X": { "shape": [1, 4096, 5120], "dtype": "float16", "dist": "normal", "seed": 612, "scale": 2 }, "bias": { "shape": [5120], "dtype": "float16", "dist": "normal", "seed": 613, "scale": 1 } }, "outputs": { "Y": { "shape": [1, 4096, 2560], "dtype": "float16" } }, "bench": { "metrics": [ { "type": "bandwidth", "value": "4096 * 5120 * dtypeBytes(args.dtype) + 4096 * 2560 * dtypeBytes(args.dtype)" } ] } }, { "name": "biassplitgelu-f16-compact-vec4-validation-gap", "preset": "edge", "vars": { "dtype": "float16" }, "inputs": { "X": { "shape": [1, 1, 8], "dtype": "float16", "dist": "normal", "seed": 614, "scale": 2 }, "bias": { "shape": [8], "dtype": "float16", "dist": "normal", "seed": 615, "scale": 1 } }, "outputs": { "Y": { "shape": [1, 1, 4], "dtype": "float16" } }, "bench": { "metrics": [{ "type": "bandwidth", "value": "8 * dtypeBytes(args.dtype) + 4 * dtypeBytes(args.dtype)" }] } }, { "name": "biassplitgelu-f32-scalar-cliff-4096x5116", "preset": "stress", "vars": { "dtype": "float32" }, "inputs": { "X": { "shape": [1, 4096, 5116], "dtype": "float32", "dist": "normal", "seed": 620, "scale": 2 }, "bias": { "shape": [5116], "dtype": "float32", "dist": "normal", "seed": 621, "scale": 1 } }, "outputs": { "Y": { "shape": [1, 4096, 2558], "dtype": "float32" } }, "bench": { "metrics": [ { "type": "bandwidth", "value": "4096 * 5116 * dtypeBytes(args.dtype) + 4096 * 2558 * dtypeBytes(args.dtype)" } ] } } ] }