{
  "id": "profiles/qwen-3-reranker-0-6b-metal-throughput",
  "name": "qwen-3-reranker-0-6b-metal-throughput",
  "description": "Qwen 3 Reranker 0.6B browser/Metal throughput profile preserving manifest compute precision while using fixed head_dim=128 f16-KV prefill attention.",
  "intent": "calibrate",
  "compatibleIntents": ["calibrate"],
  "stability": "canonical",
  "owner": "doppler-core",
  "createdAtUtc": "2026-07-09T00:00:00Z",
  "extends": "profiles/rerank-throughput",
  "runtime": {
    "inference": {
      "executionPatch": {
        "addKernels": [
          {
            "key": "attn_head128",
            "kernel": {
              "kernel": "attention_head128_f16kv.wgsl",
              "entry": "main",
              "digest": "sha256:559a3da4f3f5dcc4671f5a9b429a84209340a66a6e17f583bb329b2e23832c9b",
              "precision": {
                "kvDtype": "f16"
              }
            }
          }
        ],
        "set": [
          {
            "section": "prefill",
            "op": "attention",
            "kernelKey": "attn_head128"
          }
        ]
      },
      "session": {
        "prefillChunkLayers": 28,
        "retainQ4KMaterialization": true,
        "useWideTileQ4KPrefill": true
      }
    }
  }
}