|
Download final_report.md from GMLRVigil/BenchCheck-Pool-Frames: direct link, hf CLI and curl.
- Browser
- Download file 3.91 kB
-
https://huggingface.co/datasets/GMLRVigil/BenchCheck-Pool-Frames/resolve/main/final_report.md
- Command line
-
hf download hf://datasets/GMLRVigil/BenchCheck-Pool-Frames/final_report.md
-
curl -L -o final_report.md https://huggingface.co/datasets/GMLRVigil/BenchCheck-Pool-Frames/resolve/main/final_report.md
3.91 kB
BenchCheck-Pool final run report
Step 1
- Eligible: 299366
- Done (status=ok): 299366
- Removed: 107858
- Survived: 191508
- Pending: 0
- Permanently failed: 0
- Error rows: 2063
- Error item examples: ['ALLVB_11165', 'ALLVB_11166', 'ALLVB_11167', 'ALLVB_11168', 'ALLVB_11169', 'ALLVB_1117', 'ALLVB_11170', 'ALLVB_11171', 'ALLVB_11172', 'ALLVB_11173', 'ALLVB_11174', 'ALLVB_11175', 'ALLVB_11176', 'ALLVB_11177', 'ALLVB_11179', 'ALLVB_1118', 'ALLVB_11180', 'ALLVB_11181', 'ALLVB_11182', 'ALLVB_11183']
- Malformed rows: 0
- Successful rows without exactly 4 forwards/permutations: 0
Step 2
- Eligible: 114495
- Done (status=ok): 114495
- Removed: 36091
- Survived: 78404
- Pending: 0
- Permanently failed: 0
- Error rows: 3
- Error item examples: ['SpatialTreeBench_539', 'SpatialTreeBench_592', 'SpatialTreeBench_685']
- Malformed rows: 0
- Invalid frame/forward contract rows: 0
- Frame-count distribution: {'1': 114495}
Step 3
- Eligible: 78404
- Done (status=ok): 78404
- Removed: 22641
- Survived: 55763
- Pending: 0
- Permanently failed: 0
- Error rows: 0
- Error item examples: []
- Malformed rows: 0
- Invalid frame/forward contract rows: 0
- Frame-count distribution: {'1': 124, '2': 78, '3': 103, '4': 356, '5': 253, '6': 755, '7': 378, '8': 2820, '9': 482, '10': 2364, '11': 1029, '12': 1198, '13': 526, '14': 1074, '15': 459, '16': 2349, '17': 572, '18': 561, '19': 395, '20': 1771, '21': 899, '22': 575, '23': 255, '24': 998, '25': 246, '26': 326, '27': 251, '28': 317, '29': 234, '30': 659, '31': 263, '32': 55734}
Timing runs
- 2026-09-09 02:02:10 step1 shard 2/4: items=50902, removed=15799, errors=0, wall_s=3390.7, items_per_s=15.012
- 2026-09-09 02:02:20 step1 shard 3/4: items=50960, removed=19902, errors=0, wall_s=3400.8, items_per_s=14.985
- 2026-09-09 02:02:31 step1 shard 0/4: items=50878, removed=20050, errors=0, wall_s=3411.1, items_per_s=14.915
- 2026-09-09 02:09:41 step1 shard 1/4: items=51001, removed=18672, errors=9, wall_s=3841.0, items_per_s=13.278
- 2026-09-09 02:42:45 step1 shard 0/4: items=0, removed=0, errors=9, wall_s=17.2, items_per_s=0.0
- 2026-09-09 02:49:38 step1 shard 0/1: items=9, removed=1, errors=0, wall_s=42.4, items_per_s=0.212
- 2026-09-09 17:18:36 step2 shard 3/4: items=28629, removed=10114, errors=0, wall_s=5425.0, items_per_s=5.277
- 2026-09-09 17:18:55 step2 shard 2/4: items=28613, removed=8665, errors=0, wall_s=5444.0, items_per_s=5.256
- 2026-09-09 17:19:00 step2 shard 0/4: items=28636, removed=9185, errors=0, wall_s=5449.8, items_per_s=5.255
- 2026-09-09 17:19:39 step2 shard 1/4: items=28614, removed=8127, errors=3, wall_s=5488.0, items_per_s=5.214
- 2026-09-10 02:09:14 step3 shard 1/4: items=19596, removed=6201, errors=0, wall_s=31646.6, items_per_s=0.619
- 2026-09-10 02:13:26 step3 shard 2/4: items=19610, removed=5081, errors=0, wall_s=31898.1, items_per_s=0.615
- 2026-09-10 02:20:19 step3 shard 0/4: items=19597, removed=5480, errors=0, wall_s=32311.7, items_per_s=0.606
- 2026-09-10 02:23:51 step3 shard 3/4: items=19598, removed=5876, errors=0, wall_s=32523.0, items_per_s=0.603
- 2026-09-10 02:26:19 step2 shard 0/4: items=3, removed=0, errors=0, wall_s=22.9, items_per_s=0.131
- 2026-09-10 02:29:29 step3 shard 0/4: items=3, removed=3, errors=0, wall_s=26.6, items_per_s=0.113
Launcher changes
- Models are served from pre-downloaded local model directories while preserving served model names.
- LIBRARY_PATH includes the CUDA driver stub directory required by Triton linking on this host.
- Step 1 MAX_SEQS was lowered to 32 after startup OOM.
- Step 1 MAX_LEN was raised from 8192 to 16384 to finish nine long-context items.
- Step 2 MAX_LEN was raised from 8192 to 16384 to finish three long-context items.
- Step 3 retries use MAX_LEN 32768 so long-context items are not left unfinished.
- The launcher archives existing per-GPU logs before a resume so retry logs do not overwrite run evidence.