{
  "execution_mode": "fresh local inference; CPU operator profiler around Metal execution",
  "environment": {
    "torch": "2.14.1",
    "transformers": "5.20.2",
    "platform": "macOS-27.0.1-arm64-arm-64bit",
    "python": "3.12.8"
  },
  "model_id": "Qwen/Qwen2.5-0.5B-Instruct",
  "model_revision": "7ae557604adf67be50417f59c2c2f167def9a775",
  "source_sha256": "f09d10cf253f5c7cc013997835f56d5aced87588786efa1e575bef4bc04334da",
  "profiled_wall_ms": 55.753,
  "operators": [
    {
      "name": "aten::linear",
      "calls": 169,
      "self_cpu_us": 3648.743
    },
    {
      "name": "aten::bmm",
      "calls": 48,
      "self_cpu_us": 1436.852
    },
    {
      "name": "aten::_softmax",
      "calls": 24,
      "self_cpu_us": 687.908
    },
    {
      "name": "aten::mul",
      "calls": 245,
      "self_cpu_us": 492.195
    },
    {
      "name": "aten::add",
      "calls": 172,
      "self_cpu_us": 386.367
    },
    {
      "name": "aten::copy_",
      "calls": 196,
      "self_cpu_us": 216.708
    },
    {
      "name": "aten::cat",
      "calls": 97,
      "self_cpu_us": 187.702
    },
    {
      "name": "aten::empty",
      "calls": 298,
      "self_cpu_us": 153.942
    },
    {
      "name": "aten::flatten",
      "calls": 338,
      "self_cpu_us": 124.089
    },
    {
      "name": "aten::matmul",
      "calls": 48,
      "self_cpu_us": 117.886
    },
    {
      "name": "aten::pow",
      "calls": 49,
      "self_cpu_us": 112.667
    },
    {
      "name": "aten::_to_copy",
      "calls": 124,
      "self_cpu_us": 106.667
    }
  ],
  "limits": "CPU operator durations include dispatch and synchronization overhead, not GPU kernel duration or occupancy. Profiling changes runtime; compare unprofiled repetitions separately."
}