Koshur-OCR / beam-eval.json
KashlabsAI's picture
Release Koshur OCR v2.1 — prefix beam + train-only trigram LM
e186a72 verified
Raw
History Blame Contribute Delete
2.99 kB
{
"run_id": "ctc-v2-prefix-beam-20260729T152530Z",
"decoder": "ctc_prefix_beam",
"config": {
"beam_width": 10,
"token_topk": 12,
"lm_weight": 0.4,
"token_bonus": 1.5
},
"lm": {
"type": "character_ngram",
"order": 3,
"smoothing": 0.1,
"training_rows": 100000,
"uses_validation_text": false
},
"checkpoint_sha256": "38382437ace33611c15ddbbf718ddcd88bf25f31092883bdcab540734199648d",
"seconds": 98.94471955299377,
"greedy": {
"all": {
"rows": 920,
"reference_chars": 45907,
"edits": 10383,
"insertions": 378,
"deletions": 6859,
"substitutions": 3146,
"cer": 0.22617465745964668,
"exact": 235,
"exact_rate": 0.2554347826086957,
"nonempty": 920,
"nonempty_rate": 1.0
},
"calibration": {
"rows": 200,
"reference_chars": 9824,
"edits": 2287,
"insertions": 50,
"deletions": 1619,
"substitutions": 618,
"cer": 0.2327972312703583,
"exact": 56,
"exact_rate": 0.28,
"nonempty": 200,
"nonempty_rate": 1.0
},
"heldout": {
"rows": 720,
"reference_chars": 36083,
"edits": 8096,
"insertions": 328,
"deletions": 5240,
"substitutions": 2528,
"cer": 0.2243715877282931,
"exact": 179,
"exact_rate": 0.24861111111111112,
"nonempty": 720,
"nonempty_rate": 1.0
}
},
"beam": {
"all": {
"rows": 920,
"reference_chars": 45907,
"edits": 9803,
"insertions": 736,
"deletions": 4615,
"substitutions": 4452,
"cer": 0.2135404186725336,
"exact": 241,
"exact_rate": 0.2619565217391304,
"nonempty": 920,
"nonempty_rate": 1.0
},
"calibration": {
"rows": 200,
"reference_chars": 9824,
"edits": 2138,
"insertions": 100,
"deletions": 1081,
"substitutions": 957,
"cer": 0.21763029315960913,
"exact": 55,
"exact_rate": 0.275,
"nonempty": 200,
"nonempty_rate": 1.0
},
"heldout": {
"rows": 720,
"reference_chars": 36083,
"edits": 7665,
"insertions": 636,
"deletions": 3534,
"substitutions": 3495,
"cer": 0.21242690463653244,
"exact": 186,
"exact_rate": 0.25833333333333336,
"nonempty": 720,
"nonempty_rate": 1.0
}
},
"delta": {
"all": {
"cer_absolute": -0.012634238787113089,
"exact_rate_absolute": 0.006521739130434745,
"deletions": -2244,
"insertions": 358,
"substitutions": 1306
},
"calibration": {
"cer_absolute": -0.015166938110749179,
"exact_rate_absolute": -0.0050000000000000044,
"deletions": -538,
"insertions": 50,
"substitutions": 339
},
"heldout": {
"cer_absolute": -0.011944683091760666,
"exact_rate_absolute": 0.009722222222222243,
"deletions": -1706,
"insertions": 308,
"substitutions": 967
}
}
}