File size: 2,703 Bytes
5a6d581
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
{
  "release": "3.2.0_e2c8e65c_320k",
  "training_git_revision": "e2c8e65ccba3e388219bd7a6717eadfcfdcf28d9",
  "dataset_version": "2.8.0",
  "selected_step": 320000,
  "models": {
    "attention": {
      "run": "attention_seed12345_20261105_125044_573247",
      "checkpoint": "checkpoints/checkpoint-320000.pth",
      "checkpoint_sha256": "94de61a65471355aaf8c511e46b1017ce3ec8a999b8169ab41e7ef07fb8861dd",
      "parameters": 12421153,
      "training_git_revision": "e2c8e65ccba3e388219bd7a6717eadfcfdcf28d9",
      "data_passes": 10.113140762277984,
      "validation_metrics": {
        "Total_unweighted": 3.5557737215764105,
        "metrics/particle/matching/f1": 0.6342531442642212,
        "Regression_pt": 0.23551080994686838,
        "metrics/particle/event/met_abs_error": 14.710018157958984
      },
      "history_files": 32,
      "training_log_files": 9,
      "tensorboard_files": 10
    },
    "heptv2": {
      "run": "heptv2_seed12345_20261105_125045_324715",
      "checkpoint": "checkpoints/checkpoint-320000.pth",
      "checkpoint_sha256": "919e6e681df799294319377cf58c4b5398302724659a38eb0644fe6f3c82939a",
      "parameters": 23450849,
      "training_git_revision": "e2c8e65ccba3e388219bd7a6717eadfcfdcf28d9",
      "data_passes": 10.113140762277984,
      "validation_metrics": {
        "Total_unweighted": 3.479562263704289,
        "metrics/particle/matching/f1": 0.6352432370185852,
        "Regression_pt": 0.234293016336732,
        "metrics/particle/event/met_abs_error": 19.132198333740234
      },
      "history_files": 32,
      "training_log_files": 9,
      "tensorboard_files": 10
    }
  },
  "contents": "Final 320k checkpoints, complete histories, all resumed train/valid TensorBoard streams, configs, model kwargs, manifests, all training rank logs, and final plots. Prediction parquet files and diagnostic batch dumps are excluded.",
  "comparison_metrics": {
    "description": "Archived historical FP32 reference paired by exact input SHA256 with the saved November BF16 predictions, 512 entries per QCD/ttbar/Ztt, 1536 entries and 1530 unique events per model. Separate from mixed PU/noPU validation histories.",
    "reference": "https://huggingface.co/jpata/particleflow/tree/main/cms/v2.6.0pre1/pyg-cms_20250722_101813_274478",
    "method": "2000 paired process-stratified unique-event cluster bootstrap draws, seed 20261112. Intervals measure event sampling, not seed variation.",
    "source_study": "notebooks/studies/20261112_cms_reference_comparison",
    "limits": "Reference inference is FP32 ONNX and new predictions use BF16. One training seed. Test events also enter training validation and are not an independent final holdout."
  }
}