diff --git "a/build/webgpu/bench.json" "b/build/webgpu/bench.json" --- "a/build/webgpu/bench.json" +++ "b/build/webgpu/bench.json" @@ -1,6 +1,189 @@ { "tunableSpace": { "WORKGROUP_SIZE": [64, 128, 256] }, "cases": [ + { + "name": "qwen-image21-vae-up1-1152c-256x256-k3-bias", + "preset": "stress", + "tunables": { "IMPLICIT_MAX_M_TILES": 36, "DIRECT_IM2COL_MIN_M_TILES": 37 }, + "provenance": { + "source": "Qwen-Image 2.1 VAE decoder", + "notes": "Large materialized column buffer at the 1152-channel upsampler; the model requests implicit gather." + }, + "vars": { + "batch": 1, + "inChannels": 1152, + "outChannels": 1152, + "inH": 256, + "inW": 256, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, + "inputs": { + "x": { "shape": [1, 1152, 256, 256], "dtype": "float32", "dist": "normal", "seed": 728, "scale": 0.02 }, + "w": { "shape": [1152, 1152, 3, 3], "dtype": "float32", "dist": "normal", "seed": 729, "scale": 0.02 }, + "bias": { "shape": [1152], "dtype": "float32", "dist": "normal", "seed": 730, "scale": 0.01 } + }, + "outputs": { "y": { "shape": [1, 1152, 256, 256], "dtype": "float32" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + } + }, + { + "name": "qwen-image21-vae-up2-1152to576-256x256-k3-bias", + "preset": "stress", + "tunables": { "IMPLICIT_MAX_M_TILES": 18, "DIRECT_IM2COL_MIN_M_TILES": 19 }, + "provenance": { + "source": "Qwen-Image 2.1 VAE decoder", + "notes": "Large materialized column buffer at the reducing-channel residual entry; the model requests implicit gather." + }, + "vars": { + "batch": 1, + "inChannels": 1152, + "outChannels": 576, + "inH": 256, + "inW": 256, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, + "inputs": { + "x": { "shape": [1, 1152, 256, 256], "dtype": "float32", "dist": "normal", "seed": 725, "scale": 0.02 }, + "w": { "shape": [576, 1152, 3, 3], "dtype": "float32", "dist": "normal", "seed": 726, "scale": 0.02 }, + "bias": { "shape": [576], "dtype": "float32", "dist": "normal", "seed": 727, "scale": 0.01 } + }, + "outputs": { "y": { "shape": [1, 576, 256, 256], "dtype": "float32" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + } + }, + { + "name": "qwen-image21-vae-convout-144to4-1024x1024-k3-bias", + "preset": "stress", + "provenance": { + "source": "Qwen-Image 2.1 VAE decoder", + "notes": "Final narrow-output 3x3 convolution; the graph opts into padded implicit subgroup-matrix output rows." + }, + "vars": { + "batch": 1, + "inChannels": 144, + "outChannels": 4, + "inH": 1024, + "inW": 1024, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, + "tunables": { "SGMAT_MIN_M": 4 }, + "inputs": { + "x": { "shape": [1, 144, 1024, 1024], "dtype": "float32", "dist": "normal", "seed": 715, "scale": 0.02 }, + "w": { "shape": [4, 144, 3, 3], "dtype": "float32", "dist": "normal", "seed": 716, "scale": 0.02 }, + "bias": { "shape": [4], "dtype": "float32", "dist": "normal", "seed": 723, "scale": 0.01 } + }, + "outputs": { "y": { "shape": [1, 4, 1024, 1024], "dtype": "float32" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + } + }, + { + "name": "qwen-image21-vae-up3-288c-1024x1024-k3-bias", + "preset": "stress", + "provenance": { + "source": "Qwen-Image 2.1 VAE decoder", + "notes": "Decoder up-block 3 resample convolution; nine subgroup-matrix output-channel tiles and an im2col buffer too large for one WebGPU allocation." + }, + "vars": { + "batch": 1, + "inChannels": 288, + "outChannels": 288, + "inH": 1024, + "inW": 1024, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, + "inputs": { + "x": { "shape": [1, 288, 1024, 1024], "dtype": "float32", "dist": "normal", "seed": 715, "scale": 0.02 }, + "w": { "shape": [288, 288, 3, 3], "dtype": "float32", "dist": "normal", "seed": 716, "scale": 0.02 }, + "bias": { "shape": [288], "dtype": "float32", "dist": "normal", "seed": 717, "scale": 0.01 } + }, + "outputs": { "y": { "shape": [1, 288, 1024, 1024], "dtype": "float32" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + } + }, + { + "name": "qwen-image21-vae-up2-576c-512x512-k3-bias", + "preset": "stress", + "provenance": { + "source": "Qwen-Image 2.1 VAE decoder", + "notes": "Decoder up-block 2 resample convolution; nine 64-channel subgroup-matrix tiles and an im2col buffer too large for one WebGPU allocation." + }, + "vars": { + "batch": 1, + "inChannels": 576, + "outChannels": 576, + "inH": 512, + "inW": 512, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1] }, + "inputs": { + "x": { "shape": [1, 576, 512, 512], "dtype": "float32", "dist": "normal", "seed": 718, "scale": 0.02 }, + "w": { "shape": [576, 576, 3, 3], "dtype": "float32", "dist": "normal", "seed": 719, "scale": 0.02 }, + "bias": { "shape": [576], "dtype": "float32", "dist": "normal", "seed": 720, "scale": 0.01 } + }, + "outputs": { "y": { "shape": [1, 576, 512, 512], "dtype": "float32" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + } + }, { "name": "conv-f32-b1c16m32-32x32-k3", "preset": "smoke", @@ -140,7 +323,7 @@ "preset": "stress", "tunableSpace": { "CONV1D_TILE_M": [4, 8], "CONV1D_TILE_N": [4, 8], "CONV1D_K_TILE": [16, 32] }, "provenance": { - "source": "repository-authored benchmark", + "source": "synthetic benchmark", "notes": "Realistic long-sequence audio feature block with K and output width aligned for the materialized subgroup-matrix Conv1d path; reduced-feature tiers retain the general implicit-GEMM and direct NCW implementations." }, "vars": { "batch": 1, "inChannels": 256, "outChannels": 256, "inW": 4096, "kernelW": 7, "strideW": 1, "padW": 3 }, @@ -198,7 +381,7 @@ "name": "medical-segmentation-3d-f32-b1c32m32-32x32x32-k3-pathology", "preset": "stress", "provenance": { - "source": "repository-authored benchmark", + "source": "synthetic benchmark", "notes": "Realistic 3D medical-segmentation feature block for the direct NCDHW kernel. Each output voxel serially scans 32 input channels and a 3x3x3 kernel." }, "vars": { @@ -547,8 +730,8 @@ { "name": "conv-f32-group4-k5-64c-48x48-mid-kernel-band", "provenance": { - "source": "repository-authored benchmark", - "notes": "A grouped 5x5 convolution exercises the wide-output grouped route at an intermediate kernel size between the 3x3 and 9x9 benchmark shapes." + "source": "synthetic benchmark", + "notes": "A grouped 5x5 convolution with wide output lies between the paired 3x3 and 9x9 window sizes." }, "preset": "model", "vars": { @@ -973,10 +1156,10 @@ } }, { - "name": "manifest-gap-conv-f32-group8-k9-256c-64x63-w4-boundary", + "name": "conv-f32-group8-k9-256c-64x63-w4-boundary", "preset": "stress", "provenance": { - "source": "repository-authored benchmark", + "source": "synthetic benchmark", "notes": "A realistic grouped 9x9 convolution produces 63 columns, so the final four-lane output group contains three valid columns and uses masked tail stores." }, "vars": { @@ -1479,7 +1662,7 @@ "name": "depthanything-patchembed-3to384-518x686-k14s14", "preset": "model", "provenance": { - "notes": "Control for IMPLICIT_TILED_GATHER_MODE: stride equals kernel size, so this non-overlapping patch embed remains on the materialized im2col family." + "notes": "Measures Depth Anything's patch-embed convolution: a 14x14, stride-14 kernel over a 518x686 image with 3 input and 384 output channels." }, "tunables": { "IMPLICIT_TILED_GATHER_MODE": 1 }, "tunableSpace": { "IMPLICIT_TILED_GATHER_MODE": [0] }, @@ -2756,6 +2939,6660 @@ "source": "synthetic", "notes": "Row-tile boundary sweep: full, partial, and device-dependent complete row bands share the same reduction and spatial extent." } + }, + { + "name": "halfcols_float16_b1_c1_m512_h32_w32_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 1, + "outChannels": 512, + "inH": 32, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 1, 32, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 1, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 32, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c1_m512_h33_w33_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 1, + "outChannels": 512, + "inH": 33, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 1, 33, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 1, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 33, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c7_m512_h32_w32_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 7, + "outChannels": 512, + "inH": 32, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 7, 32, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 7, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 32, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c7_m512_h33_w33_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 7, + "outChannels": 512, + "inH": 33, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 7, 33, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 7, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 33, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c8_m512_h32_w32_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 8, + "outChannels": 512, + "inH": 32, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 8, 32, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 8, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 32, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c8_m512_h33_w33_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 8, + "outChannels": 512, + "inH": 33, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 8, 33, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 8, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 33, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c15_m512_h32_w32_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 15, + "outChannels": 512, + "inH": 32, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 15, 32, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 15, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 32, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c15_m512_h33_w33_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 15, + "outChannels": 512, + "inH": 33, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 15, 33, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 15, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 33, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c16_m512_h32_w32_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 16, + "outChannels": 512, + "inH": 32, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 16, 32, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 16, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 32, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c16_m512_h33_w33_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 16, + "outChannels": 512, + "inH": 33, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 16, 33, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 16, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 33, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c31_m512_h32_w32_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 31, + "outChannels": 512, + "inH": 32, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 31, 32, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 31, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 32, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c31_m512_h33_w33_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 31, + "outChannels": 512, + "inH": 33, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 31, 33, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 31, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 33, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m512_h32_w32_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 512, + "inH": 32, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 32, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 32, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m512_h33_w33_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 512, + "inH": 33, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 33, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 33, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c33_m512_h32_w32_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 33, + "outChannels": 512, + "inH": 32, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 33, 32, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 33, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 32, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c33_m512_h33_w33_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 33, + "outChannels": 512, + "inH": 33, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 33, 33, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 33, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 33, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c64_m512_h32_w32_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 512, + "inH": 32, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 32, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 32, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c64_m512_h33_w33_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 512, + "inH": 33, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 33, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 33, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c65_m512_h32_w32_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 65, + "outChannels": 512, + "inH": 32, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 65, 32, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 65, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 32, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c65_m512_h33_w33_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 65, + "outChannels": 512, + "inH": 33, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 65, 33, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 65, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 33, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c96_m512_h32_w32_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 96, + "outChannels": 512, + "inH": 32, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 96, 32, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 96, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 32, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c96_m512_h33_w33_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 96, + "outChannels": 512, + "inH": 33, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 96, 33, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 96, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 33, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c128_m512_h32_w32_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 128, + "outChannels": 512, + "inH": 32, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 128, 32, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 128, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 32, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c128_m512_h33_w33_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 128, + "outChannels": 512, + "inH": 33, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 128, 33, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 128, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 33, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m31_h8_w8_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 31, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [31, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [31], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 31, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m32_h8_w8_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 32, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [32, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 32, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m33_h8_w8_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 33, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [33, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [33], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 33, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m63_h8_w8_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 63, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [63, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [63], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 63, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m64_h8_w8_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 64, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [64, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 64, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m65_h8_w8_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 65, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [65, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [65], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 65, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m127_h8_w8_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 127, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [127, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [127], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 127, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m128_h8_w8_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 128, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [128, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 128, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m129_h8_w8_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 129, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [129, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [129], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 129, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m255_h8_w8_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 255, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [255, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [255], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 255, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m256_h8_w8_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 256, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [256, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 256, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m257_h8_w8_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 257, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [257, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [257], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 257, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m384_h8_w8_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 384, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m512_h8_w8_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 512, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c33_m256_h1_w1025_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 33, + "outChannels": 256, + "inH": 1, + "inW": 1025, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 33, 1, 1025], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [256, 33, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [256], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 256, 1, 1025], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c33_m384_h1_w1025_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 33, + "outChannels": 384, + "inH": 1, + "inW": 1025, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 33, 1, 1025], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 33, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [384], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 384, 1, 1025], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c33_m512_h1_w1025_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 33, + "outChannels": 512, + "inH": 1, + "inW": 1025, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 33, 1, 1025], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 33, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 1, 1025], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b2_c33_m256_h1_w1025_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 2, + "inChannels": 33, + "outChannels": 256, + "inH": 1, + "inW": 1025, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [2, 33, 1, 1025], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [256, 33, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [256], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [2, 256, 1, 1025], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b2_c33_m384_h1_w1025_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 2, + "inChannels": 33, + "outChannels": 384, + "inH": 1, + "inW": 1025, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [2, 33, 1, 1025], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 33, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [384], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [2, 384, 1, 1025], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b2_c33_m512_h1_w1025_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 2, + "inChannels": 33, + "outChannels": 512, + "inH": 1, + "inW": 1025, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [2, 33, 1, 1025], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 33, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [2, 512, 1, 1025], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float32_b1_c64_m512_h32_w32_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 512, + "inH": 32, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 32, 32], "dtype": "float32", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 64, 3, 3], "dtype": "float32", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 32, 32], "dtype": "float32" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c64_m512_h32_w32_k1_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 512, + "inH": 32, + "inW": 32, + "kernelH": 1, + "kernelW": 1, + "strideH": 1, + "strideW": 1, + "padH": 0, + "padW": 0 + }, + "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 32, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 64, 1, 1], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 32, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c512_m512_h128_w128_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 512, + "outChannels": 512, + "inH": 128, + "inW": 128, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 512, 128, 128], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 512, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 128, 128], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float32_b1_c64_m512_h32_w32_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 512, + "inH": 32, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 32, 32], "dtype": "float32", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 64, 3, 3], "dtype": "float32", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 32, 32], "dtype": "float32" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c64_m512_h32_w32_k1_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 512, + "inH": 32, + "inW": 32, + "kernelH": 1, + "kernelW": 1, + "strideH": 1, + "strideW": 1, + "padH": 0, + "padW": 0 + }, + "attrs": { "strides": [1, 1], "pads": [0, 0, 0, 0], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 32, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 64, 1, 1], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 32, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c512_m512_h128_w128_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 512, + "outChannels": 512, + "inH": 128, + "inW": 128, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 512, 128, 128], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 512, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 128, 128], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c1_m128_h14_w14_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 1, + "outChannels": 128, + "inH": 14, + "inW": 14, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 1, 14, 14], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [128, 1, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [128], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 128, 14, 14], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c8_m128_h14_w14_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 8, + "outChannels": 128, + "inH": 14, + "inW": 14, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 8, 14, 14], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [128, 8, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 128, 14, 14], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m128_h14_w14_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 128, + "inH": 14, + "inW": 14, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 14, 14], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [128, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 128, 14, 14], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c64_m128_h14_w14_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 128, + "inH": 14, + "inW": 14, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 14, 14], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [128, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 128, 14, 14], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c128_m128_h14_w14_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 128, + "outChannels": 128, + "inH": 14, + "inW": 14, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 128, 14, 14], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [128, 128, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 128, 14, 14], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b2_c8_m512_h63_w65_k3_s2_d1_bias1", + "preset": "model", + "vars": { + "batch": 2, + "inChannels": 8, + "outChannels": 512, + "inH": 63, + "inW": 65, + "kernelH": 3, + "kernelW": 3, + "strideH": 2, + "strideW": 2, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [2, 2], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [2, 8, 63, 65], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 8, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [2, 512, 32, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c33_m384_h33_w33_k3_s1_d2_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 33, + "outChannels": 384, + "inH": 33, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 2, + "padW": 2 + }, + "attrs": { "strides": [1, 1], "pads": [2, 2, 2, 2], "dilations": [2, 2] }, + "inputs": { + "x": { "shape": [1, 33, 33, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 33, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [384], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 384, 33, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c33_m512_h16_w32_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 33, + "outChannels": 512, + "inH": 16, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 33, 16, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 33, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 16, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c33_m512_h16_w32_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 33, + "outChannels": 512, + "inH": 16, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 33, 16, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 33, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 16, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c1_m512_h16_w32_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 1, + "outChannels": 512, + "inH": 16, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 1, 16, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 1, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 16, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c1_m512_h16_w32_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 1, + "outChannels": 512, + "inH": 16, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 1, 16, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 1, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 16, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c8_m512_h16_w31_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 8, + "outChannels": 512, + "inH": 16, + "inW": 31, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 8, 16, 31], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 8, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 512, 16, 31], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c8_m512_h16_w31_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 8, + "outChannels": 512, + "inH": 16, + "inW": 31, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 8, 16, 31], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [512, 8, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [512], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 512, 16, 31], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c64_m64_h64_w65_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 64, + "inH": 64, + "inW": 65, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 64, 65], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [64, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 64, 64, 65], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c64_m64_h64_w65_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 64, + "inH": 64, + "inW": 65, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 64, 65], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [64, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [64], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 64, 64, 65], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m128_h32_w65_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 128, + "inH": 32, + "inW": 65, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 32, 65], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [128, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 128, 32, 65], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c32_m128_h32_w65_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 128, + "inH": 32, + "inW": 65, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 32, 65], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [128, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [128], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 128, 32, 65], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c128_m128_h32_w65_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 128, + "outChannels": 128, + "inH": 32, + "inW": 65, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 128, 32, 65], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [128, 128, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 128, 32, 65], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c128_m128_h32_w65_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 128, + "outChannels": 128, + "inH": 32, + "inW": 65, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 128, 32, 65], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [128, 128, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [128], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 128, 32, 65], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b4_c1_m32_h16_w32_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 4, + "inChannels": 1, + "outChannels": 32, + "inH": 16, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [4, 1, 16, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [32, 1, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [4, 32, 16, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b4_c1_m32_h16_w32_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 4, + "inChannels": 1, + "outChannels": 32, + "inH": 16, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [4, 1, 16, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [32, 1, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [32], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [4, 32, 16, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b8_c1_m32_h16_w32_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 8, + "inChannels": 1, + "outChannels": 32, + "inH": 16, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [8, 1, 16, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [32, 1, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [8, 32, 16, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b8_c1_m32_h16_w32_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 8, + "inChannels": 1, + "outChannels": 32, + "inH": 16, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [8, 1, 16, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [32, 1, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [32], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [8, 32, 16, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c33_m17_h1_w1025_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 33, + "outChannels": 17, + "inH": 1, + "inW": 1025, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 33, 1, 1025], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [17, 33, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 17, 1, 1025], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c33_m17_h1_w1025_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 33, + "outChannels": 17, + "inH": 1, + "inW": 1025, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 33, 1, 1025], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [17, 33, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [17], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 17, 1, 1025], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b4_c33_m17_h1_w1025_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 4, + "inChannels": 33, + "outChannels": 17, + "inH": 1, + "inW": 1025, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [4, 33, 1, 1025], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [17, 33, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [4, 17, 1, 1025], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b4_c33_m17_h1_w1025_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 4, + "inChannels": 33, + "outChannels": 17, + "inH": 1, + "inW": 1025, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [4, 33, 1, 1025], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [17, 33, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [17], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [4, 17, 1, 1025], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b2_c32_m33_h32_w33_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 2, + "inChannels": 32, + "outChannels": 33, + "inH": 32, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [2, 32, 32, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [33, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [2, 33, 32, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b2_c32_m33_h32_w33_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 2, + "inChannels": 32, + "outChannels": 33, + "inH": 32, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [2, 32, 32, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [33, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [33], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [2, 33, 32, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b4_c32_m33_h32_w33_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 4, + "inChannels": 32, + "outChannels": 33, + "inH": 32, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [4, 32, 32, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [33, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [4, 33, 32, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b4_c32_m33_h32_w33_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 4, + "inChannels": 32, + "outChannels": 33, + "inH": 32, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [4, 32, 32, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [33, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [33], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [4, 33, 32, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c64_m129_h32_w33_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 129, + "inH": 32, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 32, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [129, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 129, 32, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b1_c64_m129_h32_w33_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 129, + "inH": 32, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 32, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [129, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [129], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [1, 129, 32, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b2_c64_m129_h32_w33_k3_s1_d1_bias0", + "preset": "model", + "vars": { + "batch": 2, + "inChannels": 64, + "outChannels": 129, + "inH": 32, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [2, 64, 32, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [129, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [2, 129, 32, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "halfcols_float16_b2_c64_m129_h32_w33_k3_s1_d1_bias1", + "preset": "model", + "vars": { + "batch": 2, + "inChannels": 64, + "outChannels": 129, + "inH": 32, + "inW": 33, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [2, 64, 32, 33], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [129, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [129], "dist": "normal", "seed": 9833, "scale": 0.1 } + }, + "outputs": { "y": { "shape": [2, 129, 32, 33], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_float16_b1_c8_m384_n64", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 8, + "outChannels": 384, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 8, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 8, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_float16_b1_c8_m384_n128", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 8, + "outChannels": 384, + "inH": 8, + "inW": 16, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 8, 8, 16], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 8, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 16], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_float16_b1_c8_m384_n256", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 8, + "outChannels": 384, + "inH": 8, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 8, 8, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 8, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_float16_b1_c8_m384_n512", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 8, + "outChannels": 384, + "inH": 8, + "inW": 64, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 8, 8, 64], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 8, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 64], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_float16_b1_c32_m384_n128", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 384, + "inH": 8, + "inW": 16, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 8, 16], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 16], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_float16_b1_c32_m384_n256", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 384, + "inH": 8, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 8, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_float16_b1_c32_m384_n512", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 32, + "outChannels": 384, + "inH": 8, + "inW": 64, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 32, 8, 64], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 32, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 64], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_float16_b1_c64_m384_n64", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_float16_b1_c64_m384_n128", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 16, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 16], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 16], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_float16_b1_c64_m384_n256", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_float16_b1_c64_m384_n512", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 64, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 64], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 64], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_float16_b1_c128_m384_n64", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 128, + "outChannels": 384, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 128, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 128, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_float16_b1_c128_m384_n128", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 128, + "outChannels": 384, + "inH": 8, + "inW": 16, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 128, 8, 16], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 128, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 16], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_float16_b1_c128_m384_n256", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 128, + "outChannels": 384, + "inH": 8, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 128, 8, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 128, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_float16_b1_c128_m384_n512", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 128, + "outChannels": 384, + "inH": 8, + "inW": 64, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 128, 8, 64], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 128, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 64], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_wg64_c64_m384_n64", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {}, + "tunables": { "WORKGROUP_SIZE": 64 } + }, + { + "name": "short_wg64_c64_m384_n128", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 16, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 16], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 16], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {}, + "tunables": { "WORKGROUP_SIZE": 64 } + }, + { + "name": "short_wg64_c64_m384_n256", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {}, + "tunables": { "WORKGROUP_SIZE": 64 } + }, + { + "name": "short_wg64_c64_m384_n512", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 64, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 64], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 64], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {}, + "tunables": { "WORKGROUP_SIZE": 64 } + }, + { + "name": "short_wg128_c64_m384_n64", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {}, + "tunables": { "WORKGROUP_SIZE": 128 } + }, + { + "name": "short_wg128_c64_m384_n128", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 16, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 16], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 16], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {}, + "tunables": { "WORKGROUP_SIZE": 128 } + }, + { + "name": "short_wg128_c64_m384_n256", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {}, + "tunables": { "WORKGROUP_SIZE": 128 } + }, + { + "name": "short_wg128_c64_m384_n512", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 64, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 64], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 64], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {}, + "tunables": { "WORKGROUP_SIZE": 128 } + }, + { + "name": "short_wg256_c64_m384_n64", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {}, + "tunables": { "WORKGROUP_SIZE": 256 } + }, + { + "name": "short_wg256_c64_m384_n128", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 16, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 16], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 16], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {}, + "tunables": { "WORKGROUP_SIZE": 256 } + }, + { + "name": "short_wg256_c64_m384_n256", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {}, + "tunables": { "WORKGROUP_SIZE": 256 } + }, + { + "name": "short_wg256_c64_m384_n512", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 64, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 64], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 64], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {}, + "tunables": { "WORKGROUP_SIZE": 256 } + }, + { + "name": "short_wg512_c64_m384_n64", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {}, + "tunables": { "WORKGROUP_SIZE": 512 } + }, + { + "name": "short_wg512_c64_m384_n128", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 16, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 16], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 16], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {}, + "tunables": { "WORKGROUP_SIZE": 512 } + }, + { + "name": "short_wg512_c64_m384_n256", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {}, + "tunables": { "WORKGROUP_SIZE": 512 } + }, + { + "name": "short_wg512_c64_m384_n512", + "preset": "model", + "vars": { + "batch": 1, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 64, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [1, 64, 8, 64], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [1, 384, 8, 64], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {}, + "tunables": { "WORKGROUP_SIZE": 512 } + }, + { + "name": "short_b2_c64_m384_n64", + "preset": "model", + "vars": { + "batch": 2, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [2, 64, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [2, 384, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_b2_c64_m384_n128", + "preset": "model", + "vars": { + "batch": 2, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 16, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [2, 64, 8, 16], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [2, 384, 8, 16], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_b2_c64_m384_n256", + "preset": "model", + "vars": { + "batch": 2, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [2, 64, 8, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [2, 384, 8, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_b4_c64_m384_n64", + "preset": "model", + "vars": { + "batch": 4, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [4, 64, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [4, 384, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_b4_c64_m384_n128", + "preset": "model", + "vars": { + "batch": 4, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 16, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [4, 64, 8, 16], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [4, 384, 8, 16], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_b4_c64_m384_n256", + "preset": "model", + "vars": { + "batch": 4, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [4, 64, 8, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [4, 384, 8, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_b8_c64_m384_n64", + "preset": "model", + "vars": { + "batch": 8, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 8, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [8, 64, 8, 8], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [8, 384, 8, 8], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_b8_c64_m384_n128", + "preset": "model", + "vars": { + "batch": 8, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 16, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [8, 64, 8, 16], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [8, 384, 8, 16], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "short_b8_c64_m384_n256", + "preset": "model", + "vars": { + "batch": 8, + "inChannels": 64, + "outChannels": 384, + "inH": 8, + "inW": 32, + "kernelH": 3, + "kernelW": 3, + "strideH": 1, + "strideW": 1, + "padH": 1, + "padW": 1 + }, + "attrs": { "strides": [1, 1], "pads": [1, 1, 1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "shape": [8, 64, 8, 32], "dtype": "float16", "dist": "normal", "seed": 356, "scale": 0.2 }, + "w": { "shape": [384, 64, 3, 3], "dtype": "float16", "dist": "normal", "seed": 357, "scale": 0.02 } + }, + "outputs": { "y": { "shape": [8, 384, 8, 32], "dtype": "float16" } }, + "bench": { + "metrics": [ + { + "type": "gflops", + "value": "2 * args.batch * args.outChannels * dim(shapes.y, 2) * dim(shapes.y, 3) * args.inChannels * args.kernelH * args.kernelW" + } + ] + }, + "tunableSpace": {} + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x32_k3x3_d1x1_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [1, 1, 1, 1], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x31_k3x3_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [1, 1, 1, 1], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x32_k3x3_d2x2_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [2, 2, 2, 2], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x31_k3x3_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [2, 2, 2, 2], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x32_k3x3_d3x3_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [3, 3, 3, 3], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x31_k3x3_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [3, 3, 3, 3], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x32_k5x5_d1x1_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [2, 2, 2, 2], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x31_k5x5_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [2, 2, 2, 2], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x32_k5x5_d2x2_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [4, 4, 4, 4], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x31_k5x5_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [4, 4, 4, 4], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x32_k5x5_d3x3_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x31_k5x5_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x32_k7x7_d1x1_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [3, 3, 3, 3], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x31_k7x7_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [3, 3, 3, 3], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x32_k7x7_d2x2_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x31_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x32_k7x7_d3x3_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [9, 9, 9, 9], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x31_k7x7_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [9, 9, 9, 9], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x32_k9x9_d1x1_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [4, 4, 4, 4], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x31_k9x9_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [4, 4, 4, 4], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x32_k9x9_d2x2_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [8, 8, 8, 8], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x31_k9x9_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [8, 8, 8, 8], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x32_k9x9_d3x3_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [12, 12, 12, 12], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x31_k9x9_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [12, 12, 12, 12], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x32_k11x11_d1x1_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [5, 5, 5, 5], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x31_k11x11_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [5, 5, 5, 5], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x32_k11x11_d2x2_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [10, 10, 10, 10], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x31_k11x11_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [10, 10, 10, 10], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x32_k11x11_d3x3_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [15, 15, 15, 15], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c8_m16_32x31_k11x11_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [15, 15, 15, 15], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [64, 8, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x32_k3x3_d1x1_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [1, 1, 1, 1], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x31_k3x3_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [1, 1, 1, 1], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x32_k3x3_d2x2_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [2, 2, 2, 2], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x31_k3x3_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [2, 2, 2, 2], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x32_k3x3_d3x3_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [3, 3, 3, 3], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x31_k3x3_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [3, 3, 3, 3], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x32_k5x5_d1x1_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [2, 2, 2, 2], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x31_k5x5_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [2, 2, 2, 2], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x32_k5x5_d2x2_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [4, 4, 4, 4], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x31_k5x5_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [4, 4, 4, 4], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x32_k5x5_d3x3_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x31_k5x5_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x32_k7x7_d1x1_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [3, 3, 3, 3], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x31_k7x7_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [3, 3, 3, 3], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x32_k7x7_d2x2_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x31_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x32_k7x7_d3x3_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [9, 9, 9, 9], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x31_k7x7_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [9, 9, 9, 9], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x32_k9x9_d1x1_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [4, 4, 4, 4], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x31_k9x9_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [4, 4, 4, 4], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x32_k9x9_d2x2_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [8, 8, 8, 8], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x31_k9x9_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [8, 8, 8, 8], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x32_k9x9_d3x3_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [12, 12, 12, 12], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x31_k9x9_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [12, 12, 12, 12], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x32_k11x11_d1x1_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [5, 5, 5, 5], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x31_k11x11_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [5, 5, 5, 5], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 8 input and 16 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x32_k11x11_d2x2_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [10, 10, 10, 10], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x31_k11x11_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [10, 10, 10, 10], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 8 input and 16 output channels per group, dilation 2x2, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x32_k11x11_d3x3_s1x1_bias0_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [15, 15, 15, 15], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 32], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 32] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 32 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c8_m16_32x31_k11x11_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [15, 15, 15, 15], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 32, 32, 31], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [64, 8, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [64], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 64, 32, 31] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 8 input and 16 output channels per group, dilation 3x3, and stride 1x1. The output has 32 rows and 31 columns per channel." + } + }, + { + "name": "grouped_float32_b2_g3_c3_m1_17x29_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 3, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [2, 9, 17, 29], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [3, 3, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [3], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [2, 3, 17, 29] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 3 input and 1 output channels per group, dilation 2x2, and stride 1x1. The output has 17 rows and 29 columns per channel." + } + }, + { + "name": "grouped_float16_b2_g3_c3_m1_17x29_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 3, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [2, 9, 17, 29], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [3, 3, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [3], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [2, 3, 17, 29] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 3 input and 1 output channels per group, dilation 2x2, and stride 1x1. The output has 17 rows and 29 columns per channel." + } + }, + { + "name": "grouped_float32_b2_g3_c3_m2_17x29_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 3, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [2, 9, 17, 29], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [6, 3, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [6], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [2, 6, 17, 29] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 3 input and 2 output channels per group, dilation 2x2, and stride 1x1. The output has 17 rows and 29 columns per channel." + } + }, + { + "name": "grouped_float16_b2_g3_c3_m2_17x29_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 3, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [2, 9, 17, 29], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [6, 3, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [6], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [2, 6, 17, 29] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 3 input and 2 output channels per group, dilation 2x2, and stride 1x1. The output has 17 rows and 29 columns per channel." + } + }, + { + "name": "grouped_float32_b2_g3_c3_m3_17x29_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 3, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [2, 9, 17, 29], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [9, 3, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [9], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [2, 9, 17, 29] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 3 input and 3 output channels per group, dilation 2x2, and stride 1x1. The output has 17 rows and 29 columns per channel." + } + }, + { + "name": "grouped_float16_b2_g3_c3_m3_17x29_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 3, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [2, 9, 17, 29], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [9, 3, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [9], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [2, 9, 17, 29] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 3 input and 3 output channels per group, dilation 2x2, and stride 1x1. The output has 17 rows and 29 columns per channel." + } + }, + { + "name": "grouped_float32_b2_g3_c3_m5_17x29_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 3, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [2, 9, 17, 29], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [15, 3, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [15], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [2, 15, 17, 29] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 3 input and 5 output channels per group, dilation 2x2, and stride 1x1. The output has 17 rows and 29 columns per channel." + } + }, + { + "name": "grouped_float16_b2_g3_c3_m5_17x29_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 3, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [2, 9, 17, 29], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [15, 3, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [15], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [2, 15, 17, 29] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 3 input and 5 output channels per group, dilation 2x2, and stride 1x1. The output has 17 rows and 29 columns per channel." + } + }, + { + "name": "grouped_float32_b2_g3_c3_m6_17x29_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 3, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [2, 9, 17, 29], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [18, 3, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [18], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [2, 18, 17, 29] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 3 input and 6 output channels per group, dilation 2x2, and stride 1x1. The output has 17 rows and 29 columns per channel." + } + }, + { + "name": "grouped_float16_b2_g3_c3_m6_17x29_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 3, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [2, 9, 17, 29], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [18, 3, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [18], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [2, 18, 17, 29] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 3 input and 6 output channels per group, dilation 2x2, and stride 1x1. The output has 17 rows and 29 columns per channel." + } + }, + { + "name": "grouped_float32_b2_g3_c3_m8_17x29_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 3, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [2, 9, 17, 29], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [24, 3, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [2, 24, 17, 29] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 3 input and 8 output channels per group, dilation 2x2, and stride 1x1. The output has 17 rows and 29 columns per channel." + } + }, + { + "name": "grouped_float16_b2_g3_c3_m8_17x29_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 3, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [2, 9, 17, 29], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [24, 3, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [2, 24, 17, 29] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 3 input and 8 output channels per group, dilation 2x2, and stride 1x1. The output has 17 rows and 29 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c3_m6_17x7_k5x5_d1x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [2, 4, 2, 4], "strides": [1, 1], "dilations": [1, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 12, 17, 7], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [24, 3, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 24, 17, 7] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 3 input and 6 output channels per group, dilation 1x2, and stride 1x1. The output has 17 rows and 7 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c3_m6_17x8_k5x5_d1x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [2, 4, 2, 4], "strides": [1, 1], "dilations": [1, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 12, 17, 8], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [24, 3, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 24, 17, 8] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 3 input and 6 output channels per group, dilation 1x2, and stride 1x1. The output has 17 rows and 8 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c3_m6_17x9_k5x5_d1x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [2, 4, 2, 4], "strides": [1, 1], "dilations": [1, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 12, 17, 9], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [24, 3, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 24, 17, 9] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 3 input and 6 output channels per group, dilation 1x2, and stride 1x1. The output has 17 rows and 9 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c3_m6_17x11_k5x5_d1x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [2, 6, 2, 6], "strides": [1, 1], "dilations": [1, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 12, 17, 11], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [24, 3, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 24, 17, 11] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 3 input and 6 output channels per group, dilation 1x3, and stride 1x1. The output has 17 rows and 11 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c3_m6_17x12_k5x5_d1x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [2, 6, 2, 6], "strides": [1, 1], "dilations": [1, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 12, 17, 12], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [24, 3, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 24, 17, 12] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 3 input and 6 output channels per group, dilation 1x3, and stride 1x1. The output has 17 rows and 12 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c3_m6_17x13_k5x5_d1x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 4, "pads": [2, 6, 2, 6], "strides": [1, 1], "dilations": [1, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 12, 17, 13], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [24, 3, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 24, 17, 13] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 3 input and 6 output channels per group, dilation 1x3, and stride 1x1. The output has 17 rows and 13 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c3_m6_33x35_k3x11_d1x2_s1x1_bias1_asym1", + "preset": "model", + "attrs": { "group": 4, "pads": [1, 10, 2, 12], "strides": [1, 1], "dilations": [1, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 12, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [24, 3, 3, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 24, 34, 37] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x11 convolution with 3 input and 6 output channels per group, dilation 1x2, and stride 1x1. The output has 34 rows and 37 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c3_m6_33x35_k11x3_d2x1_s1x1_bias1_asym1", + "preset": "model", + "attrs": { "group": 4, "pads": [10, 1, 11, 3], "strides": [1, 1], "dilations": [2, 1] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 12, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [24, 3, 11, 3], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 24, 34, 37] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x3 convolution with 3 input and 6 output channels per group, dilation 2x1, and stride 1x1. The output has 34 rows and 37 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c3_m6_33x35_k5x9_d1x3_s2x1_bias1_asym1", + "preset": "model", + "attrs": { "group": 4, "pads": [2, 12, 3, 14], "strides": [2, 1], "dilations": [1, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 12, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [24, 3, 5, 9], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 24, 17, 37] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x9 convolution with 3 input and 6 output channels per group, dilation 1x3, and stride 2x1. The output has 17 rows and 37 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c3_m6_33x35_k7x5_d2x2_s1x2_bias1_asym1", + "preset": "model", + "attrs": { "group": 4, "pads": [6, 4, 7, 6], "strides": [1, 2], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 12, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [24, 3, 7, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 24, 34, 19] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x5 convolution with 3 input and 6 output channels per group, dilation 2x2, and stride 1x2. The output has 34 rows and 19 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c3_m6_33x35_k11x11_d6x6_s1x1_bias1_asym1", + "preset": "model", + "attrs": { "group": 4, "pads": [30, 30, 31, 32], "strides": [1, 1], "dilations": [6, 6] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 12, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [24, 3, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 24, 34, 37] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 3 input and 6 output channels per group, dilation 6x6, and stride 1x1. The output has 34 rows and 37 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g4_c3_m6_33x35_k11x11_d7x7_s1x1_bias1_asym1", + "preset": "model", + "attrs": { "group": 4, "pads": [35, 35, 36, 37], "strides": [1, 1], "dilations": [7, 7] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 12, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [24, 3, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 24, 34, 37] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 3 input and 6 output channels per group, dilation 7x7, and stride 1x1. The output has 34 rows and 37 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c3_m6_33x35_k3x11_d1x2_s1x1_bias1_asym1", + "preset": "model", + "attrs": { "group": 4, "pads": [1, 10, 2, 12], "strides": [1, 1], "dilations": [1, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 12, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [24, 3, 3, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 24, 34, 37] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x11 convolution with 3 input and 6 output channels per group, dilation 1x2, and stride 1x1. The output has 34 rows and 37 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c3_m6_33x35_k11x3_d2x1_s1x1_bias1_asym1", + "preset": "model", + "attrs": { "group": 4, "pads": [10, 1, 11, 3], "strides": [1, 1], "dilations": [2, 1] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 12, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [24, 3, 11, 3], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 24, 34, 37] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x3 convolution with 3 input and 6 output channels per group, dilation 2x1, and stride 1x1. The output has 34 rows and 37 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c3_m6_33x35_k5x9_d1x3_s2x1_bias1_asym1", + "preset": "model", + "attrs": { "group": 4, "pads": [2, 12, 3, 14], "strides": [2, 1], "dilations": [1, 3] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 12, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [24, 3, 5, 9], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 24, 17, 37] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x9 convolution with 3 input and 6 output channels per group, dilation 1x3, and stride 2x1. The output has 17 rows and 37 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c3_m6_33x35_k7x5_d2x2_s1x2_bias1_asym1", + "preset": "model", + "attrs": { "group": 4, "pads": [6, 4, 7, 6], "strides": [1, 2], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 12, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [24, 3, 7, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 24, 34, 19] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x5 convolution with 3 input and 6 output channels per group, dilation 2x2, and stride 1x2. The output has 34 rows and 19 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c3_m6_33x35_k11x11_d6x6_s1x1_bias1_asym1", + "preset": "model", + "attrs": { "group": 4, "pads": [30, 30, 31, 32], "strides": [1, 1], "dilations": [6, 6] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 12, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [24, 3, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 24, 34, 37] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 3 input and 6 output channels per group, dilation 6x6, and stride 1x1. The output has 34 rows and 37 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g4_c3_m6_33x35_k11x11_d7x7_s1x1_bias1_asym1", + "preset": "model", + "attrs": { "group": 4, "pads": [35, 35, 36, 37], "strides": [1, 1], "dilations": [7, 7] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 12, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [24, 3, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [24], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 24, 34, 37] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 3 input and 6 output channels per group, dilation 7x7, and stride 1x1. The output has 34 rows and 37 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m1_33x35_k3x3_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [2, 2, 2, 2], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [16, 1, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 1 input and 1 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m2_33x35_k3x3_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [2, 2, 2, 2], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [32, 1, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 1 input and 2 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m1_33x35_k3x3_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [3, 3, 3, 3], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [16, 1, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 1 input and 1 output channels per group, dilation 3x3, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m2_33x35_k3x3_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [3, 3, 3, 3], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [32, 1, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 1 input and 2 output channels per group, dilation 3x3, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m1_33x35_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [16, 1, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 1 input and 1 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m2_33x35_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [32, 1, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 1 input and 2 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m1_33x35_k7x7_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [9, 9, 9, 9], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [16, 1, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 1 input and 1 output channels per group, dilation 3x3, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m2_33x35_k7x7_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [9, 9, 9, 9], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [32, 1, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 1 input and 2 output channels per group, dilation 3x3, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m1_33x35_k11x11_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [10, 10, 10, 10], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [16, 1, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 1 input and 1 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m2_33x35_k11x11_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [10, 10, 10, 10], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [32, 1, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 1 input and 2 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m1_33x35_k11x11_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [15, 15, 15, 15], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [16, 1, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 1 input and 1 output channels per group, dilation 3x3, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m2_33x35_k11x11_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [15, 15, 15, 15], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [32, 1, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 1 input and 2 output channels per group, dilation 3x3, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m1_33x35_k3x3_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [2, 2, 2, 2], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [16, 1, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 1 input and 1 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m2_33x35_k3x3_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [2, 2, 2, 2], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [32, 1, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 1 input and 2 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m1_33x35_k3x3_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [3, 3, 3, 3], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [16, 1, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 1 input and 1 output channels per group, dilation 3x3, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m2_33x35_k3x3_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [3, 3, 3, 3], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [32, 1, 3, 3], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 3x3 convolution with 1 input and 2 output channels per group, dilation 3x3, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m1_33x35_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [16, 1, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 1 input and 1 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m2_33x35_k7x7_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [6, 6, 6, 6], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [32, 1, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 1 input and 2 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m1_33x35_k7x7_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [9, 9, 9, 9], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [16, 1, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 1 input and 1 output channels per group, dilation 3x3, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m2_33x35_k7x7_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [9, 9, 9, 9], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [32, 1, 7, 7], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 7x7 convolution with 1 input and 2 output channels per group, dilation 3x3, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m1_33x35_k11x11_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [10, 10, 10, 10], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [16, 1, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 1 input and 1 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m2_33x35_k11x11_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [10, 10, 10, 10], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [32, 1, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 1 input and 2 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m1_33x35_k11x11_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [15, 15, 15, 15], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [16, 1, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 1 input and 1 output channels per group, dilation 3x3, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m2_33x35_k11x11_d3x3_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [15, 15, 15, 15], "strides": [1, 1], "dilations": [3, 3] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [32, 1, 11, 11], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 11x11 convolution with 1 input and 2 output channels per group, dilation 3x3, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m1_33x35_k4x4_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [1, 1, 1, 1], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [16, 1, 4, 4], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 16, 32, 34] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 4x4 convolution with 1 input and 1 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 34 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m2_33x35_k4x4_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [1, 1, 1, 1], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [32, 1, 4, 4], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 32, 32, 34] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 4x4 convolution with 1 input and 2 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 34 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m1_33x35_k4x4_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [3, 3, 3, 3], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [16, 1, 4, 4], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 4x4 convolution with 1 input and 1 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m2_33x35_k4x4_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [3, 3, 3, 3], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [32, 1, 4, 4], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 4x4 convolution with 1 input and 2 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m1_33x35_k5x5_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [2, 2, 2, 2], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [16, 1, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 1 input and 1 output channels per group, dilation 1x1, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m2_33x35_k5x5_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [2, 2, 2, 2], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [32, 1, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 1 input and 2 output channels per group, dilation 1x1, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m1_33x35_k5x5_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [4, 4, 4, 4], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [16, 1, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 1 input and 1 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m2_33x35_k5x5_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [4, 4, 4, 4], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [32, 1, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 1 input and 2 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m1_33x35_k9x9_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [4, 4, 4, 4], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [16, 1, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 1 input and 1 output channels per group, dilation 1x1, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m2_33x35_k9x9_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [4, 4, 4, 4], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [32, 1, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 1 input and 2 output channels per group, dilation 1x1, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m1_33x35_k9x9_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [8, 8, 8, 8], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [16, 1, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 1 input and 1 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float32_b1_g16_c1_m2_33x35_k9x9_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [8, 8, 8, 8], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float32", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float32", "shape": [32, 1, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float32", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float32", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 1 input and 2 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m1_33x35_k4x4_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [1, 1, 1, 1], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [16, 1, 4, 4], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 16, 32, 34] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 4x4 convolution with 1 input and 1 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 34 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m2_33x35_k4x4_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [1, 1, 1, 1], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [32, 1, 4, 4], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 32, 32, 34] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 4x4 convolution with 1 input and 2 output channels per group, dilation 1x1, and stride 1x1. The output has 32 rows and 34 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m1_33x35_k4x4_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [3, 3, 3, 3], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [16, 1, 4, 4], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 4x4 convolution with 1 input and 1 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m2_33x35_k4x4_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [3, 3, 3, 3], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [32, 1, 4, 4], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 4x4 convolution with 1 input and 2 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m1_33x35_k5x5_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [2, 2, 2, 2], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [16, 1, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 1 input and 1 output channels per group, dilation 1x1, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m2_33x35_k5x5_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [2, 2, 2, 2], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [32, 1, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 1 input and 2 output channels per group, dilation 1x1, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m1_33x35_k5x5_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [4, 4, 4, 4], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [16, 1, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 1 input and 1 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m2_33x35_k5x5_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [4, 4, 4, 4], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [32, 1, 5, 5], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 5x5 convolution with 1 input and 2 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m1_33x35_k9x9_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [4, 4, 4, 4], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [16, 1, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 1 input and 1 output channels per group, dilation 1x1, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m2_33x35_k9x9_d1x1_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [4, 4, 4, 4], "strides": [1, 1], "dilations": [1, 1] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [32, 1, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 1 input and 2 output channels per group, dilation 1x1, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m1_33x35_k9x9_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [8, 8, 8, 8], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [16, 1, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [16], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 16, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 1 input and 1 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } + }, + { + "name": "grouped_float16_b1_g16_c1_m2_33x35_k9x9_d2x2_s1x1_bias1_asym0", + "preset": "model", + "attrs": { "group": 16, "pads": [8, 8, 8, 8], "strides": [1, 1], "dilations": [2, 2] }, + "inputs": { + "x": { "dtype": "float16", "shape": [1, 16, 33, 35], "dist": "normal", "seed": 601, "scale": 0.2 }, + "w": { "dtype": "float16", "shape": [32, 1, 9, 9], "dist": "normal", "seed": 602, "scale": 0.02 }, + "bias": { "dtype": "float16", "shape": [32], "dist": "normal", "seed": 603, "scale": 0.03 } + }, + "outputs": { "y": { "dtype": "float16", "shape": [1, 32, 33, 35] } }, + "bench": { + "metrics": [ + { "type": "gflops", "value": "2 * numel(shapes.y) * dim(shapes.w, 1) * dim(shapes.w, 2) * dim(shapes.w, 3)" } + ] + }, + "provenance": { + "notes": "Grouped 9x9 convolution with 1 input and 2 output channels per group, dilation 2x2, and stride 1x1. The output has 33 rows and 35 columns per channel." + } } ] }