{
  "id": "compute/f16-batched",
  "name": "f16-batched",
  "description": "F16 activations with multi-token decode batching for lower submit/readback overhead.",
  "intent": "investigate",
  "compatibleIntents": ["investigate"],
  "stability": "canonical",
  "owner": "doppler-core",
  "createdAtUtc": "2026-02-25T00:00:00Z",
  "extends": "compute/f16-activations",
  "runtime": {
    "inference": {
      "batching": {
        "batchSize": 16,
        "stopCheckMode": "batch",
        "readbackInterval": 8,
        "readbackMode": "sequential",
        "ringTokens": 8,
        "ringStop": 8,
        "ringStaging": 8
      },
      "session": {
        "decodeLoop": {
          "batchSize": 16,
          "stopCheckMode": "batch",
          "readbackInterval": 8,
          "readbackMode": "sequential",
          "ringTokens": 8,
          "ringStop": 8,
          "ringStaging": 8
        }
      }
    }
  }
}