void.1 / shard_scheduler.json
appvoid's picture
Qwen90M step 855000; complete resumable checkpoint
124a231 verified
Raw History Blame
14.5 kB
{
"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": 77256.40416056558,
"required_pages_per_second_upper_estimate": 0.005344777299382726
},
{
"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": 9657.050520070698,
"required_pages_per_second_upper_estimate": 0.004043205390606975
},
{
"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": 19314.101040141395,
"required_pages_per_second_upper_estimate": 0.004022617899564916
},
{
"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": 28971.151560212093,
"required_pages_per_second_upper_estimate": 0.0017367762996476524
},
{
"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": 28971.151560212093,
"required_pages_per_second_upper_estimate": 0.002252379603178061
},
{
"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": 28971.151560212093,
"required_pages_per_second_upper_estimate": 0.37161082541542684
}
],
"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": 0.07687787735935103,
"producer_p95_seconds": 0.20161003999965033,
"gpu_update_seconds": 0.9543286514703385,
"producer_batches_per_second": 13.00764321738086,
"gpu_batches_per_second": 1.0478570442785045,
"producer_headroom": 12.41356661045037,
"sustainable": true,
"buffer_seconds": 60.122705042631324,
"buffer_drain_seconds": null,
"shard_target_mib": 16,
"upload_pages_per_commit": 82,
"max_pending_pages": 4096,
"generation_bytes_per_second": 446772.5461731995,
"upload_bytes_per_second": 4802594.46749234,
"upload_sustainable": true,
"logical_page_rows": 4096,
"logical_cortex_page_rows": 128,
"estimated_full_shards_per_hour": 95.8669880761813,
"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": 1.2396700680001231,
"bytes": 7470065,
"pages": 39,
"shards": 3
},
{
"seconds": 1.6093791190005504,
"bytes": 874866,
"pages": 25,
"shards": 1
},
{
"seconds": 1.1941245199996047,
"bytes": 7420221,
"pages": 39,
"shards": 3
},
{
"seconds": 1.6006398999998055,
"bytes": 9959535,
"pages": 29,
"shards": 3
},
{
"seconds": 1.2196721769996657,
"bytes": 7030636,
"pages": 39,
"shards": 2
},
{
"seconds": 1.0609669329996905,
"bytes": 7430334,
"pages": 38,
"shards": 3
},
{
"seconds": 1.0616652130001967,
"bytes": 6074717,
"pages": 35,
"shards": 2
},
{
"seconds": 1.3232113710000704,
"bytes": 5040655,
"pages": 31,
"shards": 3
},
{
"seconds": 1.5191957560000446,
"bytes": 7567101,
"pages": 34,
"shards": 3
},
{
"seconds": 1.065770677000728,
"bytes": 897206,
"pages": 26,
"shards": 1
},
{
"seconds": 1.459628982999675,
"bytes": 13449017,
"pages": 40,
"shards": 4
},
{
"seconds": 1.560673630000565,
"bytes": 10527927,
"pages": 42,
"shards": 3
},
{
"seconds": 1.384273941000174,
"bytes": 1166498,
"pages": 35,
"shards": 1
},
{
"seconds": 1.245850141000119,
"bytes": 7339507,
"pages": 30,
"shards": 3
},
{
"seconds": 1.0799733499998183,
"bytes": 6017108,
"pages": 34,
"shards": 2
},
{
"seconds": 1.2428365799996755,
"bytes": 12889604,
"pages": 40,
"shards": 4
},
{
"seconds": 1.21177975899991,
"bytes": 6176322,
"pages": 36,
"shards": 2
},
{
"seconds": 1.4487905369996952,
"bytes": 925527,
"pages": 25,
"shards": 1
},
{
"seconds": 1.6813211230000888,
"bytes": 7642707,
"pages": 39,
"shards": 3
},
{
"seconds": 1.8502533870005209,
"bytes": 10875382,
"pages": 36,
"shards": 3
},
{
"seconds": 1.247339760999239,
"bytes": 856828,
"pages": 26,
"shards": 1
},
{
"seconds": 1.4432682090000526,
"bytes": 13357057,
"pages": 39,
"shards": 4
},
{
"seconds": 1.271395830000074,
"bytes": 6214888,
"pages": 36,
"shards": 2
},
{
"seconds": 1.4851975340006902,
"bytes": 6638444,
"pages": 30,
"shards": 3
},
{
"seconds": 1.5820424499997898,
"bytes": 5864908,
"pages": 27,
"shards": 2
},
{
"seconds": 1.0337117700000817,
"bytes": 1192866,
"pages": 35,
"shards": 1
},
{
"seconds": 1.3218913410000823,
"bytes": 7396091,
"pages": 30,
"shards": 3
},
{
"seconds": 1.1367835250002827,
"bytes": 5040157,
"pages": 37,
"shards": 2
},
{
"seconds": 1.2845682920014951,
"bytes": 11638802,
"pages": 28,
"shards": 3
},
{
"seconds": 1.3405057710006076,
"bytes": 7718368,
"pages": 39,
"shards": 3
},
{
"seconds": 1.021054885999547,
"bytes": 1163402,
"pages": 34,
"shards": 1
},
{
"seconds": 1.532974641999317,
"bytes": 11431908,
"pages": 35,
"shards": 4
},
{
"seconds": 1.541119751000224,
"bytes": 6183503,
"pages": 36,
"shards": 2
},
{
"seconds": 1.4027249899991148,
"bytes": 1181995,
"pages": 34,
"shards": 1
},
{
"seconds": 1.3459127579990309,
"bytes": 13166079,
"pages": 31,
"shards": 4
},
{
"seconds": 1.166742791001525,
"bytes": 5284765,
"pages": 34,
"shards": 2
},
{
"seconds": 1.747680902999491,
"bytes": 7697898,
"pages": 38,
"shards": 3
},
{
"seconds": 0.9798488249998627,
"bytes": 6153509,
"pages": 36,
"shards": 2
},
{
"seconds": 1.0631558050008607,
"bytes": 1018193,
"pages": 28,
"shards": 2
},
{
"seconds": 1.8303398019997985,
"bytes": 12284688,
"pages": 40,
"shards": 4
},
{
"seconds": 1.182235641999796,
"bytes": 6603515,
"pages": 27,
"shards": 2
},
{
"seconds": 1.285518773000149,
"bytes": 6002809,
"pages": 36,
"shards": 2
},
{
"seconds": 1.7619018259993027,
"bytes": 7674402,
"pages": 38,
"shards": 3
},
{
"seconds": 1.6144984039983683,
"bytes": 4728422,
"pages": 30,
"shards": 2
},
{
"seconds": 1.6315553869990254,
"bytes": 5703905,
"pages": 24,
"shards": 2
},
{
"seconds": 1.11825536699871,
"bytes": 2656311,
"pages": 35,
"shards": 2
},
{
"seconds": 1.9071076060008636,
"bytes": 12136915,
"pages": 37,
"shards": 3
},
{
"seconds": 1.7720820229988021,
"bytes": 11800483,
"pages": 32,
"shards": 4
},
{
"seconds": 1.0689491579996684,
"bytes": 1154476,
"pages": 34,
"shards": 1
},
{
"seconds": 1.7027444539999124,
"bytes": 6185388,
"pages": 36,
"shards": 2
},
{
"seconds": 1.5859740610012523,
"bytes": 2563401,
"pages": 35,
"shards": 2
},
{
"seconds": 1.2778493570003775,
"bytes": 10237665,
"pages": 33,
"shards": 3
},
{
"seconds": 2.496101438000551,
"bytes": 13297971,
"pages": 33,
"shards": 4
},
{
"seconds": 1.2942262079995999,
"bytes": 1101377,
"pages": 34,
"shards": 1
},
{
"seconds": 1.815276637998977,
"bytes": 10307802,
"pages": 39,
"shards": 3
},
{
"seconds": 1.1417645319997973,
"bytes": 2370286,
"pages": 28,
"shards": 2
},
{
"seconds": 1.1171746460004215,
"bytes": 6111968,
"pages": 37,
"shards": 2
},
{
"seconds": 1.2683859379994828,
"bytes": 7618531,
"pages": 38,
"shards": 3
},
{
"seconds": 1.7501807870012271,
"bytes": 11262011,
"pages": 28,
"shards": 3
},
{
"seconds": 1.2981046400000196,
"bytes": 6199716,
"pages": 36,
"shards": 2
},
{
"seconds": 1.6966266469989932,
"bytes": 7696834,
"pages": 38,
"shards": 3
},
{
"seconds": 1.3060271180002019,
"bytes": 1241981,
"pages": 35,
"shards": 1
},
{
"seconds": 1.5251753980010108,
"bytes": 9924170,
"pages": 32,
"shards": 3
},
{
"seconds": 1.1737768090006284,
"bytes": 7614180,
"pages": 39,
"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": [
2.576500048000014,
0.012398547999964649
]
}