File size: 14,532 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": 76812.83564387822,
      "required_pages_per_second_upper_estimate": 0.005314090200177537
    },
    {
      "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": 9601.604455484778,
      "required_pages_per_second_upper_estimate": 0.004019991281210342
    },
    {
      "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": 19203.208910969555,
      "required_pages_per_second_upper_estimate": 0.003999521993480526
    },
    {
      "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": 28804.813366454335,
      "required_pages_per_second_upper_estimate": 0.0017268045789155906
    },
    {
      "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": 28804.813366454335,
      "required_pages_per_second_upper_estimate": 0.0022394475402577867
    },
    {
      "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": 28804.813366454335,
      "required_pages_per_second_upper_estimate": 0.3694772176659398
    }
  ],
  "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": 0.5553132609608156,
    "producer_p95_seconds": 0.3167180290001852,
    "gpu_update_seconds": 0.9598395812624309,
    "producer_batches_per_second": 1.800785377013647,
    "gpu_batches_per_second": 1.0418407612288172,
    "producer_headroom": 1.7284650822162877,
    "sustainable": true,
    "buffer_seconds": 60.469893619533146,
    "buffer_drain_seconds": null,
    "shard_target_mib": 16,
    "upload_pages_per_commit": 79,
    "max_pending_pages": 4096,
    "generation_bytes_per_second": 247522.86638889826,
    "upload_bytes_per_second": 3195819.317741498,
    "upload_sustainable": true,
    "logical_page_rows": 4096,
    "logical_cortex_page_rows": 128,
    "estimated_full_shards_per_hour": 53.11264508962832,
    "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": 1.5108418660020106,
      "bytes": 11046966,
      "pages": 31,
      "shards": 3
    },
    {
      "seconds": 1.4716807509976206,
      "bytes": 1094934,
      "pages": 36,
      "shards": 1
    },
    {
      "seconds": 1.608142429999134,
      "bytes": 7266564,
      "pages": 31,
      "shards": 3
    },
    {
      "seconds": 1.734035388999473,
      "bytes": 6027691,
      "pages": 29,
      "shards": 2
    },
    {
      "seconds": 1.5976459840021562,
      "bytes": 11304870,
      "pages": 44,
      "shards": 3
    },
    {
      "seconds": 1.606354307001311,
      "bytes": 7330991,
      "pages": 32,
      "shards": 3
    },
    {
      "seconds": 1.5559647560003214,
      "bytes": 6263907,
      "pages": 37,
      "shards": 2
    },
    {
      "seconds": 1.8874185050008236,
      "bytes": 2469051,
      "pages": 30,
      "shards": 2
    },
    {
      "seconds": 24.49332847400001,
      "bytes": 9957694,
      "pages": 37,
      "shards": 3
    },
    {
      "seconds": 1.9510848309982975,
      "bytes": 11992256,
      "pages": 40,
      "shards": 3
    },
    {
      "seconds": 1.68195884799934,
      "bytes": 2652603,
      "pages": 36,
      "shards": 3
    },
    {
      "seconds": 1.9732827289990382,
      "bytes": 10394345,
      "pages": 35,
      "shards": 3
    },
    {
      "seconds": 1.4147246300017287,
      "bytes": 2402371,
      "pages": 27,
      "shards": 2
    },
    {
      "seconds": 2.1062088899998344,
      "bytes": 6113756,
      "pages": 37,
      "shards": 2
    },
    {
      "seconds": 1.8720440979996056,
      "bytes": 6184562,
      "pages": 37,
      "shards": 2
    },
    {
      "seconds": 1.9332000800022797,
      "bytes": 8278775,
      "pages": 32,
      "shards": 3
    },
    {
      "seconds": 2.060914223002328,
      "bytes": 10411136,
      "pages": 38,
      "shards": 3
    },
    {
      "seconds": 1.9575544710023678,
      "bytes": 7322575,
      "pages": 31,
      "shards": 3
    },
    {
      "seconds": 1.413499946000229,
      "bytes": 1206249,
      "pages": 35,
      "shards": 1
    },
    {
      "seconds": 2.568047996999667,
      "bytes": 10001629,
      "pages": 35,
      "shards": 3
    },
    {
      "seconds": 1.5792092789997696,
      "bytes": 2645048,
      "pages": 36,
      "shards": 2
    },
    {
      "seconds": 2.121729968999716,
      "bytes": 11759509,
      "pages": 29,
      "shards": 3
    },
    {
      "seconds": 1.7052124439978797,
      "bytes": 6177071,
      "pages": 37,
      "shards": 2
    },
    {
      "seconds": 2.175897373002954,
      "bytes": 7190309,
      "pages": 36,
      "shards": 3
    },
    {
      "seconds": 1.8149941150004452,
      "bytes": 6150813,
      "pages": 37,
      "shards": 2
    },
    {
      "seconds": 1.5219117590022506,
      "bytes": 2396663,
      "pages": 28,
      "shards": 2
    },
    {
      "seconds": 1.5374675419989217,
      "bytes": 6254376,
      "pages": 36,
      "shards": 2
    },
    {
      "seconds": 1.5695442620017275,
      "bytes": 11600439,
      "pages": 29,
      "shards": 3
    },
    {
      "seconds": 1.7552413789999264,
      "bytes": 6467958,
      "pages": 38,
      "shards": 3
    },
    {
      "seconds": 1.6743759019991558,
      "bytes": 6275115,
      "pages": 37,
      "shards": 2
    },
    {
      "seconds": 2.272818253000878,
      "bytes": 7589463,
      "pages": 39,
      "shards": 3
    },
    {
      "seconds": 1.5132595840004797,
      "bytes": 5701329,
      "pages": 28,
      "shards": 2
    },
    {
      "seconds": 1.5889989900024375,
      "bytes": 6430899,
      "pages": 37,
      "shards": 2
    },
    {
      "seconds": 1.6179743659995438,
      "bytes": 8234542,
      "pages": 31,
      "shards": 3
    },
    {
      "seconds": 1.6133322549976583,
      "bytes": 6059251,
      "pages": 37,
      "shards": 2
    },
    {
      "seconds": 1.6795359120005742,
      "bytes": 5921876,
      "pages": 24,
      "shards": 2
    },
    {
      "seconds": 1.7577662710027653,
      "bytes": 6866333,
      "pages": 46,
      "shards": 3
    },
    {
      "seconds": 1.93685000899859,
      "bytes": 6023833,
      "pages": 33,
      "shards": 2
    },
    {
      "seconds": 1.8011142590003146,
      "bytes": 7499768,
      "pages": 34,
      "shards": 3
    },
    {
      "seconds": 1.6402012090002245,
      "bytes": 10865294,
      "pages": 22,
      "shards": 3
    },
    {
      "seconds": 1.7843240430011065,
      "bytes": 6397585,
      "pages": 42,
      "shards": 2
    },
    {
      "seconds": 1.4411041519997525,
      "bytes": 2727332,
      "pages": 38,
      "shards": 3
    },
    {
      "seconds": 1.913498541998706,
      "bytes": 6177731,
      "pages": 28,
      "shards": 2
    },
    {
      "seconds": 2.387397704002069,
      "bytes": 11768966,
      "pages": 40,
      "shards": 4
    },
    {
      "seconds": 1.479657814001257,
      "bytes": 836473,
      "pages": 27,
      "shards": 1
    },
    {
      "seconds": 1.9911568649986293,
      "bytes": 11961950,
      "pages": 42,
      "shards": 3
    },
    {
      "seconds": 1.958790611999575,
      "bytes": 2439502,
      "pages": 29,
      "shards": 2
    },
    {
      "seconds": 1.8466758049980854,
      "bytes": 10680095,
      "pages": 40,
      "shards": 3
    },
    {
      "seconds": 1.7479712950007524,
      "bytes": 6009903,
      "pages": 28,
      "shards": 2
    },
    {
      "seconds": 2.2005035470028815,
      "bytes": 2644353,
      "pages": 37,
      "shards": 2
    },
    {
      "seconds": 1.5131899659972987,
      "bytes": 5828306,
      "pages": 28,
      "shards": 2
    },
    {
      "seconds": 1.8049468739991426,
      "bytes": 4697774,
      "pages": 26,
      "shards": 2
    },
    {
      "seconds": 2.496297884001251,
      "bytes": 13693480,
      "pages": 42,
      "shards": 4
    },
    {
      "seconds": 1.8265930059969833,
      "bytes": 5855746,
      "pages": 36,
      "shards": 2
    },
    {
      "seconds": 1.3942693540011533,
      "bytes": 2663289,
      "pages": 34,
      "shards": 2
    },
    {
      "seconds": 1.5531271910003852,
      "bytes": 10487346,
      "pages": 37,
      "shards": 3
    },
    {
      "seconds": 1.5084796250012005,
      "bytes": 1063958,
      "pages": 32,
      "shards": 1
    },
    {
      "seconds": 1.9720517980022123,
      "bytes": 7314841,
      "pages": 27,
      "shards": 3
    },
    {
      "seconds": 1.793492888999026,
      "bytes": 11753421,
      "pages": 38,
      "shards": 3
    },
    {
      "seconds": 1.5141362069989555,
      "bytes": 6622032,
      "pages": 34,
      "shards": 3
    },
    {
      "seconds": 1.5094903019999037,
      "bytes": 6206747,
      "pages": 38,
      "shards": 2
    },
    {
      "seconds": 1.7292475489994104,
      "bytes": 6419421,
      "pages": 37,
      "shards": 2
    },
    {
      "seconds": 1.6914732460027153,
      "bytes": 2366813,
      "pages": 27,
      "shards": 2
    },
    {
      "seconds": 1.8487111699978414,
      "bytes": 10763047,
      "pages": 31,
      "shards": 3
    }
  ],
  "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": [
    3.328129581999974,
    0.10924991900014902
  ]
}