Add split definitions

splits/length_ood.json

@@ -0,0 +1,7 @@
+{
+  "split": "length_ood",
+  "seed": null,
+  "n": 0,
+  "traces": [],
+  "notes": "Composed traces with n_kernels > 5000 (launch-count tail). Placeholder filter \u2014 adjust the threshold once length-OOD traces are explicitly produced."
+}

splits/test.json

@@ -0,0 +1,2472 @@
+{
+  "split": "test",
+  "seed": 0,
+  "n": 160,
+  "traces": [
+    {
+      "path": "kernels/cutlass_fmha/_out/b2h16_q1024_d128_causal/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 68,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fmha/_out/b2h16_q512_d128_no/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 68,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fmha/_out/b2h16_q512_d64_causal/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 102,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fmha/_out/b4h8_q512_d128_causal/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 68,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fmha/_out/b8h8_q2048_d64_causal/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 102,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fmha/_out/b8h8_q512_d128_no/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 68,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fmha/_out/fa_sq2048_d64_causal/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 102,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fmha/_out/q2048/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 68,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fmha/_out/q4096/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 68,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fp8_gemm/_out/m1024_n1024_k4096/input/tensor_input.npz",
+      "motif": "cutlass_fp8_gemm",
+      "n_kernels": 201,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_fp8_gemm/_out/m2048_n2048_k8192/input/tensor_input.npz",
+      "motif": "cutlass_fp8_gemm",
+      "n_kernels": 201,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_fp8_gemm/_out/m4096_n4096_k2048/input/tensor_input.npz",
+      "motif": "cutlass_fp8_gemm",
+      "n_kernels": 201,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m1024_n1024_k8192_rM_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m1024_n1024_k8192_rM_s8_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m1024_n8192_k4096_rH_s2_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m1024_n8192_k4096_rN_s2_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m1024_n8192_k4096_rN_s4_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m2048_n2048_k8192_rN_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m2048_n4096_k8192_rN_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m4096_n2048_k2048_rH_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m4096_n2048_k2048_rM_s8_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m4096_n2048_k2048_rN_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m4096_n4096_k4096_rN_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m4096_n4096_k4096_rN_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m4096_n4096_k8192_rH_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m4096_n4096_k8192_rH_s2_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m4096_n4096_k8192_rH_s8_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m4096_n4096_k8192_rM_s4_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m4096_rN/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m8192_n1024_k4096_rN_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m8192_n1024_k4096_rN_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m8192_n1024_k4096_rN_s4_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m8192_n1024_k4096_rN_s8_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m8192_n8192_k1024_rM_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m8192_n8192_k1024_rM_s4_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m8192_n8192_k1024_rM_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m8192_n8192_k1024_rN_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m8192_n8192_k1024_rN_s8_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m8192_n8192_k8192_rH_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m8192_n8192_k8192_rN_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_grouped_gemm/_out/m512_n1024_k1024/input/tensor_input.npz",
+      "motif": "cutlass_grouped_gemm",
+      "n_kernels": 900,
+      "n_unique_kernels": 7,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_bmm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_bmm"
+    },
+    {
+      "path": "kernels/cutlass_grouped_gemm/_out/m512_n512_k512/input/tensor_input.npz",
+      "motif": "cutlass_grouped_gemm",
+      "n_kernels": 900,
+      "n_unique_kernels": 7,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_bmm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_bmm"
+    },
+    {
+      "path": "kernels/cutlass_sparse_gemm/_out/m1024_n1024_k8192/input/tensor_input.npz",
+      "motif": "cutlass_sparse_gemm",
+      "n_kernels": 607,
+      "n_unique_kernels": 6,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_sparse_gemm/_out/m2048_n2048_k8192/input/tensor_input.npz",
+      "motif": "cutlass_sparse_gemm",
+      "n_kernels": 607,
+      "n_unique_kernels": 6,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_sparse_gemm/_out/m2048_n8192_k4096/input/tensor_input.npz",
+      "motif": "cutlass_sparse_gemm",
+      "n_kernels": 607,
+      "n_unique_kernels": 6,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_sparse_gemm/_out/m8192_n8192_k8192/input/tensor_input.npz",
+      "motif": "cutlass_sparse_gemm",
+      "n_kernels": 607,
+      "n_unique_kernels": 6,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/k2048/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k1024_rH_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k1024_rM_s4_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k1024_rN_s4_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k2048_rM_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k2048_rN_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k4096_rM_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k4096_rM_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k4096_rN_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k512_rN_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n4096_k2048_rM_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n4096_k2048_rN_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n4096_k2048_rN_s4_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n4096_k2048_rN_s8_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m2048_n2048_k1024_rH_s8_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m2048_n2048_k1024_rM_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m2048_n2048_k1024_rM_s8_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m2048_n2048_k1024_rN_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m2048_n2048_k2048_rH_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m2048_n2048_k4096_rH_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m2048_n2048_k8192_rM_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m2048_n512_k4096_rH_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m2048_sw8/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m4096_n1024_k2048_rH_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m4096_n4096_k1024_rH_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m4096_n4096_k1024_rH_s8_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m4096_n4096_k1024_rN_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m4096_n4096_k2048_rH_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m4096_n4096_k2048_rM_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m4096_n4096_k2048_rN_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m4096_n4096_k4096_rM_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n2048_k4096_rM_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n2048_k4096_rN_s2_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n2048_k4096_rN_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n512_k2048_rH_s8_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n512_k2048_rM_s2_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n512_k4096_rH_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n512_k4096_rH_s4_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n512_k4096_rM_s8_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n512_k4096_rM_s8_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n512_k4096_rN_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n512_k8192_rN_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n512_k8192_rN_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/rect_mk/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/gather/_out/N4M_T128/input/tensor_input.npz",
+      "motif": "gather",
+      "n_kernels": 10001,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "memory_movement"
+      ],
+      "l2_labels": [
+        "memory_movement_gather"
+      ],
+      "dominant_l1": "memory_movement",
+      "dominant_l2": "memory_movement_gather"
+    },
+    {
+      "path": "kernels/gather/_out/N8M_T128/input/tensor_input.npz",
+      "motif": "gather",
+      "n_kernels": 10001,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "memory_movement"
+      ],
+      "l2_labels": [
+        "memory_movement_gather"
+      ],
+      "dominant_l1": "memory_movement",
+      "dominant_l2": "memory_movement_gather"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P100/dt-fp16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "loss"
+      ],
+      "l2_labels": [
+        "loss_hinge"
+      ],
+      "dominant_l1": "loss",
+      "dominant_l2": "loss_hinge"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P19/dt-bf16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_relu"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_relu"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P20/dt-bf16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_leaky_relu"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_leaky_relu"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P21/dt-bf16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_sigmoid"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_sigmoid"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P22/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_tanh"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_tanh"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P23/dt-bf16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "softmax"
+      ],
+      "l2_labels": [
+        "softmax_softmax"
+      ],
+      "dominant_l1": "softmax",
+      "dominant_l2": "softmax_softmax"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P24/dt-bf16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "softmax"
+      ],
+      "l2_labels": [
+        "softmax_log_softmax"
+      ],
+      "dominant_l1": "softmax",
+      "dominant_l2": "softmax_log_softmax"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P25/dt-fp16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 2,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_swish"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_swish"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P27/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_selu"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_selu"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P28/dt-fp16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_hardsigmoid"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_hardsigmoid"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P29/dt-fp16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_softplus"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_softplus"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P30/dt-fp16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 198,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_softsign"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_softsign"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P31/dt-bf16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_elu"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_elu"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P32/dt-bf16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_hardtanh"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_hardtanh"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P33/dt-fp16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 103,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "normalization"
+      ],
+      "l2_labels": [
+        "normalization_batchnorm"
+      ],
+      "dominant_l1": "normalization",
+      "dominant_l2": "normalization_batchnorm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P34/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 99,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "normalization"
+      ],
+      "l2_labels": [
+        "normalization_instancenorm"
+      ],
+      "dominant_l1": "normalization",
+      "dominant_l2": "normalization_instancenorm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P35/dt-fp32_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 198,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "normalization"
+      ],
+      "l2_labels": [
+        "normalization_groupnorm"
+      ],
+      "dominant_l1": "normalization",
+      "dominant_l2": "normalization_groupnorm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P36/dt-fp32_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 462,
+      "n_unique_kernels": 5,
+      "T": 512,
+      "l1_labels": [
+        "normalization"
+      ],
+      "l2_labels": [
+        "normalization_rmsnorm"
+      ],
+      "dominant_l1": "normalization",
+      "dominant_l2": "normalization_rmsnorm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P37/dt-fp16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 2,
+      "T": 512,
+      "l1_labels": [
+        "normalization"
+      ],
+      "l2_labels": [
+        "normalization_frobeniusnorm"
+      ],
+      "dominant_l1": "normalization",
+      "dominant_l2": "normalization_frobeniusnorm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P38/dt-bf16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 198,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "normalization"
+      ],
+      "l2_labels": [
+        "normalization_l1norm"
+      ],
+      "dominant_l1": "normalization",
+      "dominant_l2": "normalization_l1norm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P40/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "normalization"
+      ],
+      "l2_labels": [
+        "normalization_layernorm"
+      ],
+      "dominant_l1": "normalization",
+      "dominant_l2": "normalization_layernorm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P41/dt-bf16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "pooling"
+      ],
+      "l2_labels": [
+        "pooling_max_pool"
+      ],
+      "dominant_l1": "pooling",
+      "dominant_l2": "pooling_max_pool"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P41/dt-fp32_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "pooling"
+      ],
+      "l2_labels": [
+        "pooling_max_pool"
+      ],
+      "dominant_l1": "pooling",
+      "dominant_l2": "pooling_max_pool"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P45/dt-fp32_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "pooling"
+      ],
+      "l2_labels": [
+        "pooling_avg_pool"
+      ],
+      "dominant_l1": "pooling",
+      "dominant_l2": "pooling_avg_pool"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P46/dt-fp32_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "pooling"
+      ],
+      "l2_labels": [
+        "pooling_avg_pool"
+      ],
+      "dominant_l1": "pooling",
+      "dominant_l2": "pooling_avg_pool"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P47/dt-fp16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_sum"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_sum"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P48/dt-fp16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_mean"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_mean"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P49/dt-bf16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_max"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_max"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P5/dt-bf16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "elementwise"
+      ],
+      "l2_labels": [
+        "elementwise_scalar_multiplication"
+      ],
+      "dominant_l1": "elementwise",
+      "dominant_l2": "elementwise_scalar_multiplication"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P51/dt-fp16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_argmax"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_argmax"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P52/dt-fp16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_argmin"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_argmin"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P53/dt-fp32_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_min"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_min"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P60/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "conv"
+      ],
+      "l2_labels": [
+        "conv_conv3d_standard"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv3d_standard"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P65/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "conv"
+      ],
+      "l2_labels": [
+        "conv_convtranspose2d"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_convtranspose2d"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P77/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "conv"
+      ],
+      "l2_labels": [
+        "conv_convtranspose3d"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_convtranspose3d"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P78/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "conv"
+      ],
+      "l2_labels": [
+        "conv_convtranspose2d"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_convtranspose2d"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P79/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "conv"
+      ],
+      "l2_labels": [
+        "conv_convtranspose1d"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_convtranspose1d"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P8/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P82/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "conv"
+      ],
+      "l2_labels": [
+        "conv_conv2d_depthwise"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv2d_depthwise"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P88/dt-fp32_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 264,
+      "n_unique_kernels": 6,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_gelu"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_gelu"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P89/dt-fp32_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_cumsum"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_cumsum"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P90/dt-fp16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_cumprod"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_cumprod"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P91/dt-fp32_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 165,
+      "n_unique_kernels": 2,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_cumsum"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_cumsum"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P92/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 165,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_cumsum"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_cumsum"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P94/dt-fp32_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 198,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "loss"
+      ],
+      "l2_labels": [
+        "loss_mse"
+      ],
+      "dominant_l1": "loss",
+      "dominant_l2": "loss_mse"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P95/dt-fp16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 2,
+      "T": 512,
+      "l1_labels": [
+        "loss"
+      ],
+      "l2_labels": [
+        "loss_cross_entropy"
+      ],
+      "dominant_l1": "loss",
+      "dominant_l2": "loss_cross_entropy"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P96/dt-bf16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 2,
+      "T": 512,
+      "l1_labels": [
+        "loss"
+      ],
+      "l2_labels": [
+        "loss_huber"
+      ],
+      "dominant_l1": "loss",
+      "dominant_l2": "loss_huber"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P98/dt-fp16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 198,
+      "n_unique_kernels": 6,
+      "T": 512,
+      "l1_labels": [
+        "loss"
+      ],
+      "l2_labels": [
+        "loss_kldiv"
+      ],
+      "dominant_l1": "loss",
+      "dominant_l2": "loss_kldiv"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P99/dt-fp16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 330,
+      "n_unique_kernels": 5,
+      "T": 512,
+      "l1_labels": [
+        "loss"
+      ],
+      "l2_labels": [
+        "loss_triplet_margin"
+      ],
+      "dominant_l1": "loss",
+      "dominant_l2": "loss_triplet_margin"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P100/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 330,
+      "n_unique_kernels": 6,
+      "T": 512,
+      "l1_labels": [
+        "conv",
+        "elementwise",
+        "reduction"
+      ],
+      "l2_labels": [
+        "conv_convtranspose3d",
+        "elementwise_clamp",
+        "elementwise_div",
+        "reduction_min"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_convtranspose3d"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P18/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 726,
+      "n_unique_kernels": 12,
+      "T": 512,
+      "l1_labels": [
+        "matmul",
+        "reduction"
+      ],
+      "l2_labels": [
+        "matmul_gemm",
+        "reduction_max",
+        "reduction_sum"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P23/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 198,
+      "n_unique_kernels": 6,
+      "T": 512,
+      "l1_labels": [
+        "conv",
+        "normalization",
+        "reduction"
+      ],
+      "l2_labels": [
+        "conv_conv3d_standard",
+        "normalization_groupnorm",
+        "reduction_mean"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv3d_standard"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P35/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 264,
+      "n_unique_kernels": 7,
+      "T": 512,
+      "l1_labels": [
+        "activation",
+        "conv",
+        "elementwise",
+        "pooling"
+      ],
+      "l2_labels": [
+        "activation_hardswish",
+        "activation_mish",
+        "conv_conv2d_standard",
+        "elementwise_sub",
+        "pooling_max_pool"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv2d_standard"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P37/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 198,
+      "n_unique_kernels": 6,
+      "T": 512,
+      "l1_labels": [
+        "activation",
+        "matmul",
+        "normalization",
+        "reduction"
+      ],
+      "l2_labels": [
+        "activation_swish",
+        "matmul_gemm",
+        "normalization_groupnorm",
+        "reduction_sum"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P46/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 264,
+      "n_unique_kernels": 6,
+      "T": 512,
+      "l1_labels": [
+        "activation",
+        "conv",
+        "elementwise",
+        "pooling"
+      ],
+      "l2_labels": [
+        "activation_tanh",
+        "conv_conv2d_standard",
+        "elementwise_sub",
+        "pooling_avg_pool"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv2d_standard"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P49/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 231,
+      "n_unique_kernels": 6,
+      "T": 512,
+      "l1_labels": [
+        "activation",
+        "conv",
+        "softmax"
+      ],
+      "l2_labels": [
+        "activation_sigmoid",
+        "conv_convtranspose3d",
+        "softmax_softmax"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_convtranspose3d"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P54/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 264,
+      "n_unique_kernels": 7,
+      "T": 512,
+      "l1_labels": [
+        "activation",
+        "conv",
+        "elementwise"
+      ],
+      "l2_labels": [
+        "activation_gelu",
+        "activation_leaky_relu",
+        "conv_conv2d_standard",
+        "elementwise_mul"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv2d_standard"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P57/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 231,
+      "n_unique_kernels": 6,
+      "T": 512,
+      "l1_labels": [
+        "activation",
+        "conv"
+      ],
+      "l2_labels": [
+        "activation_hardswish",
+        "activation_relu",
+        "conv_conv2d_standard"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv2d_standard"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P64/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 462,
+      "n_unique_kernels": 12,
+      "T": 512,
+      "l1_labels": [
+        "activation",
+        "matmul"
+      ],
+      "l2_labels": [
+        "activation_leaky_relu",
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P67/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "activation",
+        "conv",
+        "pooling"
+      ],
+      "l2_labels": [
+        "activation_gelu",
+        "conv_conv2d_standard",
+        "pooling_global_avg_pool"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv2d_standard"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P82/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 198,
+      "n_unique_kernels": 5,
+      "T": 512,
+      "l1_labels": [
+        "activation",
+        "conv",
+        "elementwise",
+        "reduction"
+      ],
+      "l2_labels": [
+        "activation_tanh",
+        "conv_conv2d_standard",
+        "elementwise_bias_add",
+        "elementwise_scaling",
+        "reduction_max"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv2d_standard"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P92/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 561,
+      "n_unique_kernels": 15,
+      "T": 512,
+      "l1_labels": [
+        "activation",
+        "conv",
+        "elementwise",
+        "normalization",
+        "softmax"
+      ],
+      "l2_labels": [
+        "activation_tanh",
+        "conv_conv2d_standard",
+        "elementwise_residual_add",
+        "normalization_groupnorm",
+        "softmax_logsumexp"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv2d_standard"
+    },
+    {
+      "path": "kernels/reduction/_out/N16M_T256_I4/input/tensor_input.npz",
+      "motif": "reduction",
+      "n_kernels": 20001,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_sum"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_sum"
+    },
+    {
+      "path": "kernels/scatter/_out/N1M_B16_T512/input/tensor_input.npz",
+      "motif": "scatter",
+      "n_kernels": 20001,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "memory_movement"
+      ],
+      "l2_labels": [
+        "memory_movement_scatter"
+      ],
+      "dominant_l1": "memory_movement",
+      "dominant_l2": "memory_movement_scatter"
+    },
+    {
+      "path": "kernels/scatter/_out/N256K_B16_T128/input/tensor_input.npz",
+      "motif": "scatter",
+      "n_kernels": 20001,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "memory_movement"
+      ],
+      "l2_labels": [
+        "memory_movement_scatter"
+      ],
+      "dominant_l1": "memory_movement",
+      "dominant_l2": "memory_movement_scatter"
+    },
+    {
+      "path": "kernels/scatter/_out/N256K_B64/input/tensor_input.npz",
+      "motif": "scatter",
+      "n_kernels": 20001,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "memory_movement"
+      ],
+      "l2_labels": [
+        "memory_movement_scatter"
+      ],
+      "dominant_l1": "memory_movement",
+      "dominant_l2": "memory_movement_scatter"
+    },
+    {
+      "path": "kernels/vector_add/_out/N128M_T1024/input/tensor_input.npz",
+      "motif": "vector_add",
+      "n_kernels": 10001,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "elementwise"
+      ],
+      "l2_labels": [
+        "elementwise_add"
+      ],
+      "dominant_l1": "elementwise",
+      "dominant_l2": "elementwise_add"
+    },
+    {
+      "path": "kernels/vector_add/_out/N4M_T128/input/tensor_input.npz",
+      "motif": "vector_add",
+      "n_kernels": 10001,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "elementwise"
+      ],
+      "l2_labels": [
+        "elementwise_add"
+      ],
+      "dominant_l1": "elementwise",
+      "dominant_l2": "elementwise_add"
+    }
+  ],
+  "notes": "L2-stratified test split: ~10% of traces per L2 class (guarantees \u2265 1 entry per L1 class with \u2265 3 traces)."
+}

splits/val.json

@@ -0,0 +1,2467 @@
+{
+  "split": "val",
+  "seed": 0,
+  "n": 160,
+  "traces": [
+    {
+      "path": "kernels/cutlass_fmha/_out/b2h16_q1024_d64_causal/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 102,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fmha/_out/b2h16_q2048_d128_causal/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 68,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fmha/_out/b2h16_q512_d64_no/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 102,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fmha/_out/b4h16_q2048_d128_causal/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 68,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fmha/_out/b4h16_q512_d128_causal/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 68,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fmha/_out/b4h8_q4096_d128_causal/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 68,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fmha/_out/b4h8_q512_d128_no/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 68,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fmha/_out/b8h8_q2048_d64_no/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 102,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fmha/_out/fa_sq6144_d128/input/tensor_input.npz",
+      "motif": "cutlass_fmha",
+      "n_kernels": 68,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "attention"
+      ],
+      "l2_labels": [
+        "attention_scaled_dot_product"
+      ],
+      "dominant_l1": "attention",
+      "dominant_l2": "attention_scaled_dot_product"
+    },
+    {
+      "path": "kernels/cutlass_fp8_gemm/_out/m4096_n2048_k2048/input/tensor_input.npz",
+      "motif": "cutlass_fp8_gemm",
+      "n_kernels": 201,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_fp8_gemm/_out/m4096_n4096_k4096/input/tensor_input.npz",
+      "motif": "cutlass_fp8_gemm",
+      "n_kernels": 201,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m1024_n1024_k8192_rH_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m1024_n1024_k8192_rH_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m1024_n1024_k8192_rN_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m1024_n1024_k8192_rN_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m1024_n8192_k4096_rH_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m1024_n8192_k4096_rH_s8_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m1024_n8192_k4096_rM_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m2048/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m2048_n2048_k2048_rN_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m2048_n2048_k2048_rN_s4_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m2048_n2048_k2048_rN_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m2048_n2048_k8192_rM_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m2048_n2048_k8192_rM_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m2048_n4096_k8192_rH_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m2048_n4096_k8192_rN_s8_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m2048_rN_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m4096_n2048_k2048_rM_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m4096_n2048_k2048_rM_s2_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m4096_n4096_k4096_rH_s8_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m4096_n4096_k4096_rN_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m4096_n4096_k4096_rN_s2_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m4096_n4096_k8192_rN_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m8192_n1024_k4096_rM_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m8192_n8192_k1024_rN_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m8192_n8192_k1024_rN_s8_b1/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m8192_n8192_k8192_rH_s4_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_gemm/_out/m8192_n8192_k8192_rM_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_gemm",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_grouped_gemm/_out/m2048_n2048_k512/input/tensor_input.npz",
+      "motif": "cutlass_grouped_gemm",
+      "n_kernels": 900,
+      "n_unique_kernels": 7,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_bmm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_bmm"
+    },
+    {
+      "path": "kernels/cutlass_grouped_gemm/_out/m2048_n512_k1024/input/tensor_input.npz",
+      "motif": "cutlass_grouped_gemm",
+      "n_kernels": 900,
+      "n_unique_kernels": 7,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_bmm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_bmm"
+    },
+    {
+      "path": "kernels/cutlass_sparse_gemm/_out/m1024_n2048_k4096/input/tensor_input.npz",
+      "motif": "cutlass_sparse_gemm",
+      "n_kernels": 607,
+      "n_unique_kernels": 6,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_sparse_gemm/_out/m2048_n1024_k8192/input/tensor_input.npz",
+      "motif": "cutlass_sparse_gemm",
+      "n_kernels": 607,
+      "n_unique_kernels": 6,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/k1024/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k1024_rH_s4_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k1024_rM_s8_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k1024_rN_s8_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k1024_rN_s8_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k4096_rM_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k4096_rM_s8_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k512_rH_s4_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k512_rM_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k512_rM_s8_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k512_rN_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k512_rN_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k8192_rH_s8_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k8192_rM_s4_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k8192_rM_s8_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n1024_k8192_rN_s8_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n4096_k2048_rH_s2_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n4096_k2048_rH_s4_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n4096_k2048_rM_s4_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n4096_k2048_rN_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_n4096_k2048_rN_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m1024_sw4/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m2048_n2048_k2048_rH_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m2048_n2048_k2048_rH_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m2048_n2048_k2048_rM_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m2048_n2048_k8192_rH_s2_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m2048_n2048_k8192_rH_s8_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m2048_n2048_k8192_rM_s8_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m2048_n512_k4096_rM_s8_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m2048_rM/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m4096_n1024_k2048_rH_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m4096_n1024_k2048_rM_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m4096_n4096_k1024_rM_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m4096_n4096_k2048_rH_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m4096_n4096_k2048_rM_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m4096_n4096_k2048_rM_s2_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m4096_n4096_k4096_rH_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m4096_n4096_k4096_rM_s1_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m4096_n4096_k4096_rN_s8_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n2048_k4096_rH_s2_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n2048_k4096_rH_s8_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n2048_k4096_rM_s4_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n2048_k4096_rM_s8_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n2048_k4096_rM_s8_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n2048_k4096_rN_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n512_k2048_rN_s1_b1/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n512_k8192_rN_s4_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/cutlass_ws_overlap/_out/m512_n512_k8192_rN_s8_b0/input/tensor_input.npz",
+      "motif": "cutlass_ws_overlap",
+      "n_kernels": 206,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/gather/_out/N16M_T128/input/tensor_input.npz",
+      "motif": "gather",
+      "n_kernels": 10001,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "memory_movement"
+      ],
+      "l2_labels": [
+        "memory_movement_gather"
+      ],
+      "dominant_l1": "memory_movement",
+      "dominant_l2": "memory_movement_gather"
+    },
+    {
+      "path": "kernels/gather/_out/N256K_T128/input/tensor_input.npz",
+      "motif": "gather",
+      "n_kernels": 10001,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "memory_movement"
+      ],
+      "l2_labels": [
+        "memory_movement_gather"
+      ],
+      "dominant_l1": "memory_movement",
+      "dominant_l2": "memory_movement_gather"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P100/dt-bf16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "loss"
+      ],
+      "l2_labels": [
+        "loss_hinge"
+      ],
+      "dominant_l1": "loss",
+      "dominant_l2": "loss_hinge"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P19/dt-fp32_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_relu"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_relu"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P20/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_leaky_relu"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_leaky_relu"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P21/dt-fp32_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_sigmoid"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_sigmoid"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P22/dt-fp32_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_tanh"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_tanh"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P23/dt-fp16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "softmax"
+      ],
+      "l2_labels": [
+        "softmax_softmax"
+      ],
+      "dominant_l1": "softmax",
+      "dominant_l2": "softmax_softmax"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P24/dt-fp16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "softmax"
+      ],
+      "l2_labels": [
+        "softmax_log_softmax"
+      ],
+      "dominant_l1": "softmax",
+      "dominant_l2": "softmax_log_softmax"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P25/dt-fp32_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 264,
+      "n_unique_kernels": 2,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_swish"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_swish"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P27/dt-fp32_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_selu"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_selu"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P28/dt-bf16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_hardsigmoid"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_hardsigmoid"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P29/dt-bf16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_softplus"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_softplus"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P30/dt-fp16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 198,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_softsign"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_softsign"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P31/dt-bf16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_elu"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_elu"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P32/dt-bf16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_hardtanh"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_hardtanh"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P33/dt-fp16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 103,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "normalization"
+      ],
+      "l2_labels": [
+        "normalization_batchnorm"
+      ],
+      "dominant_l1": "normalization",
+      "dominant_l2": "normalization_batchnorm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P34/dt-fp32_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 99,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "normalization"
+      ],
+      "l2_labels": [
+        "normalization_instancenorm"
+      ],
+      "dominant_l1": "normalization",
+      "dominant_l2": "normalization_instancenorm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P35/dt-fp16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 134,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "normalization"
+      ],
+      "l2_labels": [
+        "normalization_groupnorm"
+      ],
+      "dominant_l1": "normalization",
+      "dominant_l2": "normalization_groupnorm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P36/dt-fp16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 264,
+      "n_unique_kernels": 5,
+      "T": 512,
+      "l1_labels": [
+        "normalization"
+      ],
+      "l2_labels": [
+        "normalization_rmsnorm"
+      ],
+      "dominant_l1": "normalization",
+      "dominant_l2": "normalization_rmsnorm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P37/dt-fp32_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 264,
+      "n_unique_kernels": 2,
+      "T": 512,
+      "l1_labels": [
+        "normalization"
+      ],
+      "l2_labels": [
+        "normalization_frobeniusnorm"
+      ],
+      "dominant_l1": "normalization",
+      "dominant_l2": "normalization_frobeniusnorm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P38/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 396,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "normalization"
+      ],
+      "l2_labels": [
+        "normalization_l1norm"
+      ],
+      "dominant_l1": "normalization",
+      "dominant_l2": "normalization_l1norm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P40/dt-fp16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 35,
+      "n_unique_kernels": 2,
+      "T": 512,
+      "l1_labels": [
+        "normalization"
+      ],
+      "l2_labels": [
+        "normalization_layernorm"
+      ],
+      "dominant_l1": "normalization",
+      "dominant_l2": "normalization_layernorm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P41/dt-fp16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "pooling"
+      ],
+      "l2_labels": [
+        "pooling_max_pool"
+      ],
+      "dominant_l1": "pooling",
+      "dominant_l2": "pooling_max_pool"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "pooling"
+      ],
+      "l2_labels": [
+        "pooling_max_pool"
+      ],
+      "dominant_l1": "pooling",
+      "dominant_l2": "pooling_max_pool"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P44/dt-fp32_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "pooling"
+      ],
+      "l2_labels": [
+        "pooling_avg_pool"
+      ],
+      "dominant_l1": "pooling",
+      "dominant_l2": "pooling_avg_pool"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P46/dt-fp32_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "pooling"
+      ],
+      "l2_labels": [
+        "pooling_avg_pool"
+      ],
+      "dominant_l1": "pooling",
+      "dominant_l2": "pooling_avg_pool"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P48/dt-fp32_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_mean"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_mean"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P49/dt-fp16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_max"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_max"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P5/dt-bf16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "elementwise"
+      ],
+      "l2_labels": [
+        "elementwise_scalar_multiplication"
+      ],
+      "dominant_l1": "elementwise",
+      "dominant_l2": "elementwise_scalar_multiplication"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P51/dt-fp32_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_argmax"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_argmax"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P52/dt-bf16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_argmin"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_argmin"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P53/dt-bf16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_min"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_min"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P62/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "conv"
+      ],
+      "l2_labels": [
+        "conv_conv2d_standard"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv2d_standard"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P63/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 99,
+      "n_unique_kernels": 2,
+      "T": 512,
+      "l1_labels": [
+        "conv"
+      ],
+      "l2_labels": [
+        "conv_conv2d_standard"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv2d_standard"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P64/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "conv"
+      ],
+      "l2_labels": [
+        "conv_convtranspose1d"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_convtranspose1d"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P69/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "conv"
+      ],
+      "l2_labels": [
+        "conv_convtranspose2d"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_convtranspose2d"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P76/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 99,
+      "n_unique_kernels": 2,
+      "T": 512,
+      "l1_labels": [
+        "conv"
+      ],
+      "l2_labels": [
+        "conv_conv1d_standard"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv1d_standard"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P83/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "conv"
+      ],
+      "l2_labels": [
+        "conv_conv2d_depthwise"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv2d_depthwise"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P88/dt-bf16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 264,
+      "n_unique_kernels": 6,
+      "T": 512,
+      "l1_labels": [
+        "activation"
+      ],
+      "l2_labels": [
+        "activation_gelu"
+      ],
+      "dominant_l1": "activation",
+      "dominant_l2": "activation_gelu"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P89/dt-fp16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_cumsum"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_cumsum"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P9/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "matmul"
+      ],
+      "l2_labels": [
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P90/dt-bf16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 33,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_cumprod"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_cumprod"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P92/dt-bf16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_cumsum"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_cumsum"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P93/dt-fp16_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 2,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_cumsum"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_cumsum"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P94/dt-fp16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 99,
+      "n_unique_kernels": 3,
+      "T": 512,
+      "l1_labels": [
+        "loss"
+      ],
+      "l2_labels": [
+        "loss_mse"
+      ],
+      "dominant_l1": "loss",
+      "dominant_l2": "loss_mse"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P95/dt-bf16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 2,
+      "T": 512,
+      "l1_labels": [
+        "loss"
+      ],
+      "l2_labels": [
+        "loss_cross_entropy"
+      ],
+      "dominant_l1": "loss",
+      "dominant_l2": "loss_cross_entropy"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P96/dt-fp16_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 66,
+      "n_unique_kernels": 2,
+      "T": 512,
+      "l1_labels": [
+        "loss"
+      ],
+      "l2_labels": [
+        "loss_huber"
+      ],
+      "dominant_l1": "loss",
+      "dominant_l2": "loss_huber"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P98/dt-fp32_s123/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 198,
+      "n_unique_kernels": 6,
+      "T": 512,
+      "l1_labels": [
+        "loss"
+      ],
+      "l2_labels": [
+        "loss_kldiv"
+      ],
+      "dominant_l1": "loss",
+      "dominant_l2": "loss_kldiv"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L1_P99/dt-fp32_s42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 330,
+      "n_unique_kernels": 5,
+      "T": 512,
+      "l1_labels": [
+        "loss"
+      ],
+      "l2_labels": [
+        "loss_triplet_margin"
+      ],
+      "dominant_l1": "loss",
+      "dominant_l2": "loss_triplet_margin"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P17/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 264,
+      "n_unique_kernels": 7,
+      "T": 512,
+      "l1_labels": [
+        "conv",
+        "elementwise",
+        "normalization"
+      ],
+      "l2_labels": [
+        "conv_conv2d_standard",
+        "elementwise_div",
+        "normalization_instancenorm"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv2d_standard"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P20/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 297,
+      "n_unique_kernels": 6,
+      "T": 512,
+      "l1_labels": [
+        "conv",
+        "elementwise",
+        "reduction"
+      ],
+      "l2_labels": [
+        "conv_convtranspose3d",
+        "elementwise_mul",
+        "elementwise_residual_add",
+        "reduction_sum"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_convtranspose3d"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P24/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "conv",
+        "reduction",
+        "softmax"
+      ],
+      "l2_labels": [
+        "conv_conv3d_standard",
+        "reduction_min",
+        "softmax_softmax"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv3d_standard"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P26/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 264,
+      "n_unique_kernels": 7,
+      "T": 512,
+      "l1_labels": [
+        "activation",
+        "conv",
+        "elementwise"
+      ],
+      "l2_labels": [
+        "activation_hardswish",
+        "conv_convtranspose3d",
+        "elementwise_add"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_convtranspose3d"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P32/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 198,
+      "n_unique_kernels": 5,
+      "T": 512,
+      "l1_labels": [
+        "conv",
+        "elementwise",
+        "reduction"
+      ],
+      "l2_labels": [
+        "conv_conv2d_standard",
+        "elementwise_scaling",
+        "reduction_min"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv2d_standard"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P42/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 660,
+      "n_unique_kernels": 15,
+      "T": 512,
+      "l1_labels": [
+        "conv",
+        "elementwise",
+        "pooling"
+      ],
+      "l2_labels": [
+        "conv_convtranspose2d",
+        "elementwise_bias_add",
+        "elementwise_mul",
+        "pooling_global_avg_pool"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_convtranspose2d"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P45/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 396,
+      "n_unique_kernels": 11,
+      "T": 512,
+      "l1_labels": [
+        "matmul",
+        "softmax"
+      ],
+      "l2_labels": [
+        "matmul_gemm",
+        "softmax_logsumexp"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P48/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 198,
+      "n_unique_kernels": 5,
+      "T": 512,
+      "l1_labels": [
+        "activation",
+        "conv",
+        "elementwise"
+      ],
+      "l2_labels": [
+        "activation_sigmoid",
+        "activation_tanh",
+        "conv_conv3d_standard",
+        "elementwise_mul",
+        "elementwise_scaling"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv3d_standard"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P70/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 132,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "activation",
+        "elementwise",
+        "matmul"
+      ],
+      "l2_labels": [
+        "activation_sigmoid",
+        "elementwise_scaling",
+        "matmul_gemm"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P75/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 165,
+      "n_unique_kernels": 5,
+      "T": 512,
+      "l1_labels": [
+        "matmul",
+        "normalization",
+        "reduction"
+      ],
+      "l2_labels": [
+        "matmul_gemm",
+        "normalization_groupnorm",
+        "reduction_min"
+      ],
+      "dominant_l1": "matmul",
+      "dominant_l2": "matmul_gemm"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P87/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 165,
+      "n_unique_kernels": 4,
+      "T": 512,
+      "l1_labels": [
+        "activation",
+        "conv",
+        "elementwise"
+      ],
+      "l2_labels": [
+        "activation_mish",
+        "conv_conv2d_standard",
+        "elementwise_sub"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv2d_standard"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P90/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 198,
+      "n_unique_kernels": 5,
+      "T": 512,
+      "l1_labels": [
+        "activation",
+        "conv",
+        "elementwise",
+        "reduction"
+      ],
+      "l2_labels": [
+        "activation_gelu",
+        "activation_leaky_relu",
+        "conv_conv3d_standard",
+        "elementwise_clamp",
+        "reduction_sum"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_conv3d_standard"
+    },
+    {
+      "path": "kernels/kernelbench/_out/L2_P96/input/tensor_input.npz",
+      "motif": "kernelbench",
+      "n_kernels": 297,
+      "n_unique_kernels": 9,
+      "T": 512,
+      "l1_labels": [
+        "conv",
+        "elementwise",
+        "pooling",
+        "reduction"
+      ],
+      "l2_labels": [
+        "conv_convtranspose3d",
+        "elementwise_mul",
+        "pooling_global_avg_pool",
+        "reduction_max"
+      ],
+      "dominant_l1": "conv",
+      "dominant_l2": "conv_convtranspose3d"
+    },
+    {
+      "path": "kernels/reduction/_out/N16M_T512_I4/input/tensor_input.npz",
+      "motif": "reduction",
+      "n_kernels": 20001,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_sum"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_sum"
+    },
+    {
+      "path": "kernels/reduction/_out/N1M_T256_I4/input/tensor_input.npz",
+      "motif": "reduction",
+      "n_kernels": 20001,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "reduction"
+      ],
+      "l2_labels": [
+        "reduction_sum"
+      ],
+      "dominant_l1": "reduction",
+      "dominant_l2": "reduction_sum"
+    },
+    {
+      "path": "kernels/scatter/_out/N256K_B256_T128/input/tensor_input.npz",
+      "motif": "scatter",
+      "n_kernels": 20001,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "memory_movement"
+      ],
+      "l2_labels": [
+        "memory_movement_scatter"
+      ],
+      "dominant_l1": "memory_movement",
+      "dominant_l2": "memory_movement_scatter"
+    },
+    {
+      "path": "kernels/scatter/_out/N256K_B256_T512/input/tensor_input.npz",
+      "motif": "scatter",
+      "n_kernels": 20001,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "memory_movement"
+      ],
+      "l2_labels": [
+        "memory_movement_scatter"
+      ],
+      "dominant_l1": "memory_movement",
+      "dominant_l2": "memory_movement_scatter"
+    },
+    {
+      "path": "kernels/scatter/_out/N4M_B1024_T128/input/tensor_input.npz",
+      "motif": "scatter",
+      "n_kernels": 20001,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "memory_movement"
+      ],
+      "l2_labels": [
+        "memory_movement_scatter"
+      ],
+      "dominant_l1": "memory_movement",
+      "dominant_l2": "memory_movement_scatter"
+    },
+    {
+      "path": "kernels/vector_add/_out/N1M_T128/input/tensor_input.npz",
+      "motif": "vector_add",
+      "n_kernels": 10001,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "elementwise"
+      ],
+      "l2_labels": [
+        "elementwise_add"
+      ],
+      "dominant_l1": "elementwise",
+      "dominant_l2": "elementwise_add"
+    },
+    {
+      "path": "kernels/vector_add/_out/N1M_T512/input/tensor_input.npz",
+      "motif": "vector_add",
+      "n_kernels": 10001,
+      "n_unique_kernels": 1,
+      "T": 512,
+      "l1_labels": [
+        "elementwise"
+      ],
+      "l2_labels": [
+        "elementwise_add"
+      ],
+      "dominant_l1": "elementwise",
+      "dominant_l2": "elementwise_add"
+    }
+  ],
+  "notes": "L2-stratified val split: ~10% of traces per L2 class (guarantees \u2265 1 entry per L1 class with \u2265 2 traces in the full record set)."
+}