Add sampled_id_list.json (33-definition dtype-symmetric subset) and document it in the dataset card

README.md

@@ -43,6 +43,7 @@ solutions/     # kernels implementing a definition
     reference-scalar/conv2d/  # naive scalar 6-loop conv2d (ground-truth correctness)
 workloads/     # concrete problem instances (var-axis values + scalar inputs)
   conv/        #   one JSONL per definition; each line is a runnable workload
+sampled_id_list.json  # 33-definition representative subset (see below)
 ```
 
 ### Definitions (`definitions/conv/*.json`)
@@ -97,6 +98,39 @@ The shared **harness** (`solutions/ncnn/_harness/conv2d.{h,cpp}`) declares the
 `ncnn::convolution_kernel` contract that every solution implements, and handles
 input padding before the kernel is invoked.
 
+## Sampled subset (`sampled_id_list.json`)
+
+The full trace is 103 definitions. `sampled_id_list.json` names a
+**33-definition representative subset**, for runs where a full sweep is too
+expensive:
+
+```json
+{
+  "description": "33-definition dtype-symmetric representative sample of bench-trace/definitions/",
+  "count": 33,
+  "definitions": [
+    {"name": "conv2d_fp32_kh3_kw3_sh1_sw1_dh1_dw1_p1", "dataset": "ncnn"},
+    {"name": "gemm_bf16_n2048_k2048", "dataset": "llama.cpp"},
+    {"name": "loop_001", "dataset": "simd-loop"}
+  ]
+}
+```
+
+| Dataset     | Sampled | Total |
+|-------------|--------:|------:|
+| ncnn        |      12 |    30 |
+| llama.cpp   |       8 |    26 |
+| simd-loop   |      13 |    47 |
+
+**Dtype-symmetric** means each sampled op type appears in *every* dtype this
+dataset provides for it, on identical shapes — so comparing dtypes within one
+op type is never confounded by which shapes happen to have been sampled:
+
+- `conv2d`, `conv2d_depthwise`, `gemm` (ncnn) — fp32 + w8a8ch
+- `gemm`, `moe` (llama.cpp) — bf16 + q8_0 + q4_k_m
+- `mha` (bf16-only), `pooling`, `rms_norm` (fp32-only) — single-dtype op types
+  here, so one entry each
+
 ## Intended use
 
 Drive an ARM CPU kernel benchmark/agent loop:

sampled_id_list.json

@@ -0,0 +1,138 @@
+{
+  "description": "33-definition dtype-symmetric representative sample of bench-trace/definitions/",
+  "count": 33,
+  "definitions": [
+    {
+      "name": "conv2d_depthwise_fp32_kh3_kw3_sh1_sw1_dh1_dw1_p1",
+      "dataset": "ncnn"
+    },
+    {
+      "name": "conv2d_depthwise_fp32_kh5_kw5_sh1_sw1_dh1_dw1_p2",
+      "dataset": "ncnn"
+    },
+    {
+      "name": "conv2d_depthwise_w8a8ch_kh3_kw3_sh1_sw1_dh1_dw1_p1",
+      "dataset": "ncnn"
+    },
+    {
+      "name": "conv2d_depthwise_w8a8ch_kh5_kw5_sh1_sw1_dh1_dw1_p2",
+      "dataset": "ncnn"
+    },
+    {
+      "name": "conv2d_fp32_kh3_kw3_sh1_sw1_dh1_dw1_p1",
+      "dataset": "ncnn"
+    },
+    {
+      "name": "conv2d_fp32_kh7_kw7_sh2_sw2_dh1_dw1_p3",
+      "dataset": "ncnn"
+    },
+    {
+      "name": "conv2d_w8a8ch_kh3_kw3_sh1_sw1_dh1_dw1_p1",
+      "dataset": "ncnn"
+    },
+    {
+      "name": "conv2d_w8a8ch_kh7_kw7_sh2_sw2_dh1_dw1_p3",
+      "dataset": "ncnn"
+    },
+    {
+      "name": "gemm_fp32_n1000_k2048",
+      "dataset": "ncnn"
+    },
+    {
+      "name": "gemm_w8a8ch_n1000_k2048",
+      "dataset": "ncnn"
+    },
+    {
+      "name": "pooling_fp32_global_avg",
+      "dataset": "ncnn"
+    },
+    {
+      "name": "pooling_fp32_max_kh3_kw3_sh2_sw2_p1",
+      "dataset": "ncnn"
+    },
+    {
+      "name": "gemm_bf16_n2048_k2048",
+      "dataset": "llama.cpp"
+    },
+    {
+      "name": "gemm_q4_k_m_n2048_k2048",
+      "dataset": "llama.cpp"
+    },
+    {
+      "name": "gemm_q8_0_n2048_k2048",
+      "dataset": "llama.cpp"
+    },
+    {
+      "name": "mha_bf16_h16_d128_kvh16",
+      "dataset": "llama.cpp"
+    },
+    {
+      "name": "moe_bf16_e64_k8_d2048_ff1024",
+      "dataset": "llama.cpp"
+    },
+    {
+      "name": "moe_q4_k_m_e64_k8_d2048_ff1024",
+      "dataset": "llama.cpp"
+    },
+    {
+      "name": "moe_q8_0_e64_k8_d2048_ff1024",
+      "dataset": "llama.cpp"
+    },
+    {
+      "name": "rms_norm_fp32_d2048",
+      "dataset": "llama.cpp"
+    },
+    {
+      "name": "loop_001",
+      "dataset": "simd-loop"
+    },
+    {
+      "name": "loop_008",
+      "dataset": "simd-loop"
+    },
+    {
+      "name": "loop_024",
+      "dataset": "simd-loop"
+    },
+    {
+      "name": "loop_037",
+      "dataset": "simd-loop"
+    },
+    {
+      "name": "loop_102",
+      "dataset": "simd-loop"
+    },
+    {
+      "name": "loop_106",
+      "dataset": "simd-loop"
+    },
+    {
+      "name": "loop_121",
+      "dataset": "simd-loop"
+    },
+    {
+      "name": "loop_123",
+      "dataset": "simd-loop"
+    },
+    {
+      "name": "loop_127",
+      "dataset": "simd-loop"
+    },
+    {
+      "name": "loop_128",
+      "dataset": "simd-loop"
+    },
+    {
+      "name": "loop_130",
+      "dataset": "simd-loop"
+    },
+    {
+      "name": "loop_135",
+      "dataset": "simd-loop"
+    },
+    {
+      "name": "loop_223",
+      "dataset": "simd-loop"
+    }
+  ]
+}