File size: 14,536 Bytes

{
  "version": 1,
  "startup_batch_seconds": [
    2.3260779750000324,
    0.017838069999982054,
    0.0158013189999906,
    0.015721255000016754,
    0.09197544800002788,
    0.014327330000014626,
    0.013824727999974584,
    0.10075297500003444
  ],
  "source_benchmarks": [
    {
      "source": "climbmix",
      "rows": 256,
      "tokens": 903410,
      "bytes": 372598,
      "generation_seconds": 1.5044533369999726,
      "write_seconds": 0.005470204000005197,
      "read_seconds": 0.004841493000014907,
      "tokens_per_second": 598315.063954628,
      "note": "Bounded fresh sample; full-page latency is an extrapolation.",
      "estimated_pages_for_buffer": 1,
      "required_token_positions_per_second": 76793.98366392785,
      "required_pages_per_second_upper_estimate": 0.005312785976461951
    },
    {
      "source": "rewrite6",
      "rows": 256,
      "tokens": 149279,
      "bytes": 66870,
      "generation_seconds": 17.489590872999997,
      "write_seconds": 0.0027630569999814725,
      "read_seconds": 0.003109784999992371,
      "tokens_per_second": 8533.957213384612,
      "note": "Bounded fresh sample; full-page latency is an extrapolation.",
      "estimated_pages_for_buffer": 1,
      "required_token_positions_per_second": 9599.247957990981,
      "required_pages_per_second_upper_estimate": 0.004019004664918953
    },
    {
      "source": "openmath",
      "rows": 256,
      "tokens": 300086,
      "bytes": 97823,
      "generation_seconds": 0.7871533299999669,
      "write_seconds": 0.0022467860000006112,
      "read_seconds": 0.0028803160000165917,
      "tokens_per_second": 380144.35761751566,
      "note": "Bounded fresh sample; full-page latency is an extrapolation.",
      "estimated_pages_for_buffer": 1,
      "required_token_positions_per_second": 19198.495915981963,
      "required_pages_per_second_upper_estimate": 0.003998540400914647
    },
    {
      "source": "ultra_qa",
      "rows": 256,
      "tokens": 1042562,
      "bytes": 369139,
      "generation_seconds": 23.827993708999998,
      "write_seconds": 0.0062629279999555365,
      "read_seconds": 0.006445001999964006,
      "tokens_per_second": 43742.16556775435,
      "note": "Bounded fresh sample; full-page latency is an extrapolation.",
      "estimated_pages_for_buffer": 1,
      "required_token_positions_per_second": 28797.743873972948,
      "required_pages_per_second_upper_estimate": 0.0017263807736358214
    },
    {
      "source": "ultra_style",
      "rows": 256,
      "tokens": 803904,
      "bytes": 337820,
      "generation_seconds": 36.50458967899999,
      "write_seconds": 0.00519779599994763,
      "read_seconds": 0.005391688000031536,
      "tokens_per_second": 22018.86276523721,
      "note": "Bounded fresh sample; full-page latency is an extrapolation.",
      "estimated_pages_for_buffer": 1,
      "required_token_positions_per_second": 28797.743873972948,
      "required_pages_per_second_upper_estimate": 0.00223889791831277
    },
    {
      "source": "cortex",
      "rows": 128,
      "tokens": 77961,
      "bytes": 43256,
      "generation_seconds": 0.08065045199998622,
      "write_seconds": 0.004991313999994418,
      "read_seconds": 0.04061180899998362,
      "tokens_per_second": 910315.184299418,
      "note": "Bounded fresh sample; full-page latency is an extrapolation.",
      "estimated_pages_for_buffer": 23,
      "required_token_positions_per_second": 28797.743873972948,
      "required_pages_per_second_upper_estimate": 0.36938653780701824
    }
  ],
  "gpu_measurement": "Selected forward/backward estimate; replaced by live successful update timings.",
  "layout": "Logical pages unchanged; multiple pages per immutable TAR shard.",
  "candidate_shard_mib": [
    16,
    32,
    64,
    128,
    256
  ],
  "local_shard_size_benchmark": [
    {
      "target_mib": 16,
      "actual_bytes": 15554560,
      "pages": 7,
      "packing_seconds": 0.03195600199990167,
      "read_seconds": 0.0013301719999390116,
      "io_bytes_per_second": 467297923.7588089,
      "target_reached": true
    },
    {
      "target_mib": 32,
      "actual_bytes": 18083840,
      "pages": 9,
      "packing_seconds": 0.0402841879999869,
      "read_seconds": 0.0016377959999545055,
      "io_bytes_per_second": 431368897.0451703,
      "target_reached": false
    },
    {
      "target_mib": 64,
      "actual_bytes": 18083840,
      "pages": 9,
      "packing_seconds": 0.0399535819999528,
      "read_seconds": 0.0013251859999172666,
      "io_bytes_per_second": 438090594.17802685,
      "target_reached": false
    },
    {
      "target_mib": 128,
      "actual_bytes": 18083840,
      "pages": 9,
      "packing_seconds": 0.04405787199993938,
      "read_seconds": 0.0014645830000290516,
      "io_bytes_per_second": 397250983.05907583,
      "target_reached": false
    },
    {
      "target_mib": 256,
      "actual_bytes": 18083840,
      "pages": 9,
      "packing_seconds": 0.038781767000045875,
      "read_seconds": 0.0013218610000649278,
      "io_bytes_per_second": 450927781.39548963,
      "target_reached": false
    }
  ],
  "shard_benchmark_scope": "Local packing/read measurements; upload capacity measured only on real data commits. Targets larger than available pages are sample-limited.",
  "estimated_fresh_generation_batch_seconds": 2.558590814491007,
  "selected": {
    "prefetch_batches": 63,
    "prefetch_capacity_limit": 512,
    "batch_memory_estimate_bytes": 720750,
    "producer_batch_seconds": 0.17096050562497567,
    "producer_p95_seconds": 0.25604570699942997,
    "gpu_update_seconds": 0.9600752100926881,
    "producer_batches_per_second": 5.849304179022677,
    "gpu_batches_per_second": 1.0415850648861742,
    "producer_headroom": 5.6157719385712355,
    "sustainable": true,
    "buffer_seconds": 60.48473823583935,
    "buffer_drain_seconds": null,
    "shard_target_mib": 16,
    "upload_pages_per_commit": 81,
    "max_pending_pages": 4096,
    "generation_bytes_per_second": 534534.0209854754,
    "upload_bytes_per_second": 5286647.891867109,
    "upload_sustainable": true,
    "logical_page_rows": 4096,
    "logical_cortex_page_rows": 128,
    "estimated_full_shards_per_hour": 114.69855758832165,
    "selection_policy": "Smallest shard within 90% of measured local peak, enlarged to amortize 30 seconds of measured page production; bounded to 256 MiB."
  },
  "upload_measurements": [
    {
      "seconds": 1.1387970109999515,
      "bytes": 10453428,
      "pages": 40,
      "shards": 3
    },
    {
      "seconds": 1.3777619700003925,
      "bytes": 819264,
      "pages": 26,
      "shards": 2
    },
    {
      "seconds": 1.2043475289992784,
      "bytes": 13122126,
      "pages": 40,
      "shards": 4
    },
    {
      "seconds": 1.1971378379994349,
      "bytes": 5971358,
      "pages": 36,
      "shards": 2
    },
    {
      "seconds": 0.9979249650004931,
      "bytes": 6639528,
      "pages": 32,
      "shards": 3
    },
    {
      "seconds": 1.0738694510000641,
      "bytes": 6116466,
      "pages": 37,
      "shards": 2
    },
    {
      "seconds": 0.993601779000528,
      "bytes": 1093689,
      "pages": 34,
      "shards": 1
    },
    {
      "seconds": 1.0732885580000584,
      "bytes": 7199351,
      "pages": 29,
      "shards": 3
    },
    {
      "seconds": 1.3326483809996716,
      "bytes": 16444186,
      "pages": 37,
      "shards": 4
    },
    {
      "seconds": 0.9647154859994771,
      "bytes": 2573191,
      "pages": 37,
      "shards": 2
    },
    {
      "seconds": 1.2671719289992325,
      "bytes": 6070484,
      "pages": 36,
      "shards": 2
    },
    {
      "seconds": 0.9431054610013234,
      "bytes": 842654,
      "pages": 26,
      "shards": 1
    },
    {
      "seconds": 1.3556453979999787,
      "bytes": 11128099,
      "pages": 32,
      "shards": 4
    },
    {
      "seconds": 1.1991763389996777,
      "bytes": 5989028,
      "pages": 39,
      "shards": 2
    },
    {
      "seconds": 1.483354228999815,
      "bytes": 13516314,
      "pages": 40,
      "shards": 4
    },
    {
      "seconds": 0.855230520999612,
      "bytes": 838138,
      "pages": 29,
      "shards": 1
    },
    {
      "seconds": 1.1489447340009065,
      "bytes": 10403004,
      "pages": 36,
      "shards": 3
    },
    {
      "seconds": 0.9511843500004034,
      "bytes": 2374318,
      "pages": 29,
      "shards": 2
    },
    {
      "seconds": 1.678345239999544,
      "bytes": 6141170,
      "pages": 36,
      "shards": 2
    },
    {
      "seconds": 1.0750775129999965,
      "bytes": 1235188,
      "pages": 35,
      "shards": 1
    },
    {
      "seconds": 1.2133236099998612,
      "bytes": 11087663,
      "pages": 9,
      "shards": 3
    },
    {
      "seconds": 1.060693833998812,
      "bytes": 7602919,
      "pages": 45,
      "shards": 3
    },
    {
      "seconds": 1.202748789000907,
      "bytes": 6204281,
      "pages": 36,
      "shards": 2
    },
    {
      "seconds": 1.4798138580008526,
      "bytes": 7393031,
      "pages": 30,
      "shards": 3
    },
    {
      "seconds": 0.8912479129994608,
      "bytes": 1116344,
      "pages": 34,
      "shards": 1
    },
    {
      "seconds": 1.331704049000109,
      "bytes": 10313766,
      "pages": 37,
      "shards": 3
    },
    {
      "seconds": 1.5044031500001438,
      "bytes": 8199304,
      "pages": 34,
      "shards": 3
    },
    {
      "seconds": 1.4808365860008053,
      "bytes": 6083362,
      "pages": 37,
      "shards": 2
    },
    {
      "seconds": 1.2456822759995703,
      "bytes": 12102575,
      "pages": 32,
      "shards": 4
    },
    {
      "seconds": 0.872640274999867,
      "bytes": 1185922,
      "pages": 35,
      "shards": 1
    },
    {
      "seconds": 1.1524197270009608,
      "bytes": 6081827,
      "pages": 36,
      "shards": 2
    },
    {
      "seconds": 1.1447629150006833,
      "bytes": 7321171,
      "pages": 30,
      "shards": 3
    },
    {
      "seconds": 1.0708932610014017,
      "bytes": 6812845,
      "pages": 32,
      "shards": 3
    },
    {
      "seconds": 1.235942547000377,
      "bytes": 11771801,
      "pages": 40,
      "shards": 4
    },
    {
      "seconds": 1.195207675000347,
      "bytes": 1134238,
      "pages": 34,
      "shards": 1
    },
    {
      "seconds": 1.0769621990002634,
      "bytes": 6468284,
      "pages": 36,
      "shards": 2
    },
    {
      "seconds": 2.3685029429998394,
      "bytes": 11410659,
      "pages": 34,
      "shards": 4
    },
    {
      "seconds": 0.8315555410008528,
      "bytes": 1017212,
      "pages": 30,
      "shards": 1
    },
    {
      "seconds": 1.2100102580006933,
      "bytes": 5596380,
      "pages": 21,
      "shards": 2
    },
    {
      "seconds": 1.2646026780003012,
      "bytes": 13750693,
      "pages": 48,
      "shards": 4
    },
    {
      "seconds": 1.3428173240008618,
      "bytes": 5558514,
      "pages": 31,
      "shards": 2
    },
    {
      "seconds": 1.273476110000047,
      "bytes": 7479055,
      "pages": 38,
      "shards": 3
    },
    {
      "seconds": 1.084104961999401,
      "bytes": 947822,
      "pages": 26,
      "shards": 1
    },
    {
      "seconds": 1.1121349749992078,
      "bytes": 6275461,
      "pages": 37,
      "shards": 2
    },
    {
      "seconds": 1.2975029250010266,
      "bytes": 6264701,
      "pages": 26,
      "shards": 3
    },
    {
      "seconds": 1.4136529820007127,
      "bytes": 11858220,
      "pages": 37,
      "shards": 3
    },
    {
      "seconds": 1.0470604990005086,
      "bytes": 6246532,
      "pages": 37,
      "shards": 2
    },
    {
      "seconds": 1.0089250809996884,
      "bytes": 2580280,
      "pages": 37,
      "shards": 2
    },
    {
      "seconds": 1.1785594149987446,
      "bytes": 10228291,
      "pages": 34,
      "shards": 3
    },
    {
      "seconds": 0.9172461540001677,
      "bytes": 2354642,
      "pages": 28,
      "shards": 2
    },
    {
      "seconds": 1.1485085579988663,
      "bytes": 12167168,
      "pages": 40,
      "shards": 3
    },
    {
      "seconds": 1.0151017729986052,
      "bytes": 6037243,
      "pages": 35,
      "shards": 2
    },
    {
      "seconds": 1.2390309599995817,
      "bytes": 6886865,
      "pages": 35,
      "shards": 3
    },
    {
      "seconds": 1.3525706070013257,
      "bytes": 5675921,
      "pages": 28,
      "shards": 2
    },
    {
      "seconds": 1.2551854249995813,
      "bytes": 7559192,
      "pages": 38,
      "shards": 3
    },
    {
      "seconds": 0.825367109000581,
      "bytes": 1141341,
      "pages": 33,
      "shards": 1
    },
    {
      "seconds": 1.5828043500005151,
      "bytes": 15814359,
      "pages": 33,
      "shards": 4
    },
    {
      "seconds": 1.3585308149995399,
      "bytes": 7486305,
      "pages": 37,
      "shards": 3
    },
    {
      "seconds": 0.8041142309994029,
      "bytes": 1148050,
      "pages": 34,
      "shards": 1
    },
    {
      "seconds": 1.0195367809992604,
      "bytes": 6121053,
      "pages": 30,
      "shards": 2
    },
    {
      "seconds": 2.6576393979994464,
      "bytes": 1839915,
      "pages": 12,
      "shards": 2
    },
    {
      "seconds": 1.339241851999759,
      "bytes": 10779600,
      "pages": 40,
      "shards": 3
    },
    {
      "seconds": 4.318171066001014,
      "bytes": 1094935,
      "pages": 34,
      "shards": 1
    },
    {
      "seconds": 1.3115034270012984,
      "bytes": 13301880,
      "pages": 34,
      "shards": 5
    }
  ],
  "cache_key": {
    "version": 1,
    "kind": "data",
    "gpu": {
      "name": "NVIDIA RTX PRO 6000 Blackwell Server Edition",
      "total_memory": 101975851008,
      "capability": [
        12,
        0
      ]
    },
    "torch": "2.8.0+cu128",
    "cuda": "12.8",
    "python": "3.13.11",
    "packages": {
      "transformers": "4.57.1",
      "numpy": "2.2.6",
      "pyarrow": "21.0.0",
      "huggingface_hub": "0.36.0"
    },
    "settings": {
      "context": 2048,
      "global_batch": 90,
      "page_rows": 4096,
      "cortex_page_rows": 128,
      "mix": {
        "climbmix": 40,
        "ultra_qa": 15,
        "ultra_style": 15,
        "cortex": 15,
        "openmath": 10,
        "rewrite6": 5
      },
      "cpu_threads": 4
    },
    "source": "f8722b66ff2f92bca174f36564336149e85c65c48b964d05ff3a4d5d85da308f",
    "cpu": "unknown",
    "cpu_count": 20,
    "microbatch": 90
  },
  "cache_reused": true,
  "resume_probe_seconds": [
    3.0803219919999947,
    0.10207374499998423
  ]
}