{
  "recorded_at": "2026-08-26",
  "updated_at": "2026-08-26T06:13:05+01:00",
  "host": {
    "name": "evox3",
    "system": "GMKtec EVO-X3",
    "processor": "AMD Ryzen AI MAX+ 395",
    "gpu": "Radeon 8060S",
    "memory_gib": 128
  },
  "model": {
    "name": "Ornith-1.5-35B-A3B-Q6_K",
    "filename": "Ornith-1.5-35B-Q6_K.gguf",
    "size_bytes": 29208731200,
    "sha256": "1c4e5bb98a74c89a5d93a2488b5748b7b331daf77f6dbb64bd9b2ff864b64eb3"
  },
  "runtime": {
    "llama_cpp_release": "b10641",
    "source_commit": "28a3e60e5b0b162bec0c35527ea7247d00543fea",
    "server_sha256": "debeaafb8cf09d64ea3d98cb5f6aa616e4150b6b03e1bd5618660c2775e27df9",
    "production_safety_patch_sha256": "c04fee55c166a25fe696dea22ca1dce96fdfe43d496fb29e8a5ba8bc318fb396",
    "vulkan_driver": "Strix Halo llama.cpp v0.6.11",
    "lemonade_version": "11.7.0"
  },
  "profile": {
    "public_aliases": [
      "ornith",
      "ornith35",
      "Ornith-1.5-35B-A3B-Q6_K"
    ],
    "public_port": 13305,
    "backend_binding": "loopback-only, Lemonade-managed",
    "context_tokens": 32768,
    "batch_tokens": 2048,
    "ubatch_tokens": 2048,
    "kv_key_type": "q8_0",
    "kv_value_type": "q8_0",
    "flash_attention": true,
    "parallel_slots": 1,
    "host_prompt_cache_mib": 8192,
    "cache_prompt": true,
    "cache_idle_slots": true,
    "speculation": {
      "type": "draft-mtp",
      "adaptive": true,
      "draft_n_min": 2,
      "draft_n_max": 4,
      "draft_p_min": 0
    },
    "auto_evict": true,
    "downsize_idle_seconds": 120,
    "evict_idle_seconds": 900,
    "evict_weight_factor": 2.0
  },
  "qualification": {
    "prefill_refresh": {
      "run_id": "20260826T055054+0100",
      "launch_order": [
        "current-b10635",
        "patched-b10641",
        "patched-b10641",
        "current-b10635"
      ],
      "launches_per_build": 2,
      "repetitions_per_launch_cell": 5,
      "samples_per_build_cell": 10,
      "benchmark_records": 24,
      "raw_samples": 120,
      "summary_statistic": "median across ten raw observations per build/depth/ubatch cell",
      "gpu_fault_lines": 0,
      "production_service_restored": true,
      "current_llama_bench_sha256": "c66acb56b32c4924da0eafa54b1a30b6c95b4d96bdc95b6c785c731ff2e5763f",
      "patched_llama_bench_sha256": "9aa1ca56899705003703c2f9af3e9059c6c3de9d0730997016a6748f76d0dbaa",
      "vulkan_driver_sha256": "56947038530a3ba1b5a006b7d598fc42170a04aa666639dd6f825e6628d8453f",
      "data_files": {
        "ornith15-b10641-prefill-2026-08-26.csv": "66793994eccccbe64298b405f1359fc6ab64af82a1211a79891c2acf371d88b3",
        "ornith15-b10641-prefill-summary-2026-08-26.csv": "aca989a329ac9b5395d0ff7c6814620c0874d3a1b312a803e03c4917e48d39eb",
        "ornith15-b10641-prefill-raw-2026-08-26.jsonl": "4413aa26fe9001028b09ec2db71e523166df598ccf485bc890b09e26748c2210",
        "ornith15-b10641-prefill-analysis-2026-08-26.json": "8476b9d1970f1fdff5b18380d20f9c776d0ee854ba491c36bc0a44b9a2a197c7"
      }
    },
    "direct_soak_seconds": 1807.389692916011,
    "lemonade_soak_seconds": 1806.2348956020141,
    "lemonade_soak_passed": true,
    "server_error_lines": 0,
    "gpu_fault_lines": 0,
    "production_overlap": false
  },
  "boundaries": {
    "qualified_context_tokens": 32768,
    "model_architectural_context_tokens": 262144,
    "not_tested": [
      "64K or longer context with ubatch 2048",
      "multi-slot serving",
      "adaptive MTP against fixed MTP as a controlled performance A/B",
      "power consumption"
    ]
  }
}