| { | |
| "model": "arabic-clitic-tokenizer", | |
| "metric_scale": "percent 0-100", | |
| "tokenizer_historical_training_record": { | |
| "code": "ar", | |
| "language": "Arabic", | |
| "component": "tokenizer", | |
| "split": "development", | |
| "checkpoint": "training/trankit_save_ja_ner_v2/xlm-roberta-base/arabic/arabic.tokenizer.mdl", | |
| "checkpoint_sha256": "fd2a9195aabc1a39f98e339969732c61d37cf994c648f6d8867db0554591bc7f", | |
| "checkpoint_epoch": 11, | |
| "matched_training_output": "_deprecated_nonruntime_20260514_184819/training/t_ar10k2/xlm-roberta-base/customized-mwt/customized-mwt.tokenizer.mdl", | |
| "log": "_deprecated_nonruntime_20260514_184819/training/t_ar10k2/xlm-roberta-base/customized-mwt/logs/tokenize.training", | |
| "log_sha256": "64b853045f27875f2c2ae4560deb97cc613e87054c8ef9a16424d3c4fed5aa2a", | |
| "selected_table_line": 9513, | |
| "selected_metrics_f1_percent": { | |
| "Tokens": 99.86, | |
| "Sentences": 97.49 | |
| }, | |
| "last_logged_metrics_f1_percent": { | |
| "Tokens": 99.86, | |
| "Sentences": 97.49 | |
| }, | |
| "selection_evidence": "Checkpoint stores this zero-based epoch; the corresponding completed-epoch table is selected by ordinal, not by the misleading repeated best-epoch header." | |
| }, | |
| "mwt_development_reevaluation": { | |
| "language_code": "ar", | |
| "language": "Arabic", | |
| "checkpoint": { | |
| "path": "training/trankit_save_ja_ner_v2/xlm-roberta-base/arabic/arabic_mwt_expander.pt", | |
| "sha256": "c9e2cd69fa05a44e6b16d56112b1fcec74654bba7e6998e0d9c2877c2273c044", | |
| "bytes": 1384901 | |
| }, | |
| "evaluation_date": "2026-09-24", | |
| "device": "CPU", | |
| "checkpoint_format": "native PyTorch, unquantized", | |
| "split": "retained development material", | |
| "input": { | |
| "path": "_deprecated_nonruntime_20260514_184819/training/t_ar10k2/xlm-roberta-base/customized-mwt/preds/tokenizer.dev.conllu", | |
| "sha256": "83be526273aeb88f058564bfc02569e011c24ff53bf22b022a64c80fa452628c", | |
| "bytes": 428935 | |
| }, | |
| "input_kind": "retained_tokenizer_development_predictions", | |
| "reference": { | |
| "path": "_deprecated_nonruntime_20260514_184819/training/arabiccomponents/lessbad/ara_news_2022_10K_plain_trankit_tok_dev_fixed.conllu", | |
| "sha256": "7f142595a3f116ab557c07230a0e447cb7b31799358676e366796e2db186d12c", | |
| "bytes": 806800 | |
| }, | |
| "sentence_count": 497, | |
| "mwt_candidate_count": 1962, | |
| "dictionary_ensemble": { | |
| "precision": 98.108346, | |
| "recall": 98.071366, | |
| "f1": 98.089852 | |
| }, | |
| "seq2seq_only": { | |
| "precision": 97.806687, | |
| "recall": 97.76982, | |
| "f1": 97.78825 | |
| }, | |
| "prediction_sha256": "77eb44b3c8e233c532ab0c520d124c0f502424188b7d3c6247198a92668e9386", | |
| "torch_version": "2.12.0.dev20260217+cu128", | |
| "scope_note": "Retained selected-tokenizer development predictions feed the expander. The input token/sentence metrics match the selected tokenizer training log." | |
| } | |
| } | |