2026-05-29-crf-classweights-clean

This model is a fine-tuned version of Davlan/bert-base-multilingual-cased-ner-hrl on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 26.0250
• Precision: 0.7676
• Recall: 0.8088
• F1: 0.7877
• Accuracy: 0.9662
• Academicdiscipline F1: 0.6667
• Ambiguouslydefinedconcept F1: 0.8383
• Discoursephenomenon F1: 0.7154
• Graphemicphenomenon F1: 0.0
• Languagerelatedterm F1: 0.8106
• Languageresourceinformation F1: 0.7816
• Lexicalphenomenon F1: 0.6877
• Morphologicalphenomenon F1: 0.8057
• Morphosyntacticphenomenon F1: 0.8595
• New Tag F1: 0.8218
• Otherlinguisticterm F1: 0.7259
• Phonologicalphenomenon F1: 0.8404
• Semanticphenomenon F1: 0.6452
• Syntacticphenomenon F1: 0.7594
• Topnode Dummy F1: 0.6845
• Unclassifiedlinguisticconcept F1: 0.8623

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

Trained on an annotated dataset , with academic publications in the Bibliography of Linguistic Literature (BLL), labelled by human annotator (c-ho/2026-05-28_ub_opus_bll_ner_annotation).

========== LABEL DISTRIBUTION ==========

B-AcademicDiscipline | 40 | 0.000103 B-AmbiguouslyDefinedConcept | 253 | 0.000649 B-DiscoursePhenomenon | 199 | 0.000511 B-GraphemicPhenomenon | 13 | 0.000033 B-LanguageRelatedTerm | 2293 | 0.005886 B-LanguageResourceInformation | 1207 | 0.003098 B-LexicalPhenomenon | 581 | 0.001491 B-MorphologicalPhenomenon | 1393 | 0.003576 B-MorphosyntacticPhenomenon | 5665 | 0.014542 B-NEW_TAG | 3249 | 0.008340 B-OtherLinguisticTerm | 2338 | 0.006002 B-PhonologicalPhenomenon | 3424 | 0.008790 B-SemanticPhenomenon | 900 | 0.002310 B-SyntacticPhenomenon | 2765 | 0.007098 B-TOPNODE_DUMMY | 4769 | 0.012242 B-UnclassifiedLinguisticConcept | 274 | 0.000703 I-AcademicDiscipline | 30 | 0.000077 I-AmbiguouslyDefinedConcept | 2 | 0.000005 I-DiscoursePhenomenon | 29 | 0.000074 I-GraphemicPhenomenon | 5 | 0.000013 I-LanguageRelatedTerm | 136 | 0.000349 I-LanguageResourceInformation | 73 | 0.000187 I-LexicalPhenomenon | 59 | 0.000151 I-MorphologicalPhenomenon | 23 | 0.000059 I-MorphosyntacticPhenomenon | 354 | 0.000909 I-NEW_TAG | 124 | 0.000318 I-OtherLinguisticTerm | 112 | 0.000288 I-PhonologicalPhenomenon | 206 | 0.000529 I-SemanticPhenomenon | 86 | 0.000221 I-SyntacticPhenomenon | 252 | 0.000647 I-TOPNODE_DUMMY | 525 | 0.001348 I-UnclassifiedLinguisticConcept | 23 | 0.000059 O | 358151 | 0.919390

========================================

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 5e-05
• train_batch_size: 16
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 0.1
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy	Academicdiscipline F1	Ambiguouslydefinedconcept F1	Discoursephenomenon F1	Graphemicphenomenon F1	Languagerelatedterm F1	Languageresourceinformation F1	Lexicalphenomenon F1	Morphologicalphenomenon F1	Morphosyntacticphenomenon F1	New Tag F1	Otherlinguisticterm F1	Phonologicalphenomenon F1	Semanticphenomenon F1	Syntacticphenomenon F1	Topnode Dummy F1	Unclassifiedlinguisticconcept F1
No log	1.0	196	88.7338	0.0	0.0	0.0	0.9155	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0	0.0
No log	2.0	392	27.9599	0.6174	0.6089	0.6131	0.9460	0.0	0.0698	0.0	0.0	0.6775	0.5802	0.4863	0.7394	0.7285	0.6741	0.5367	0.7704	0.2400	0.5667	0.3722	0.2692
467.1807	3.0	588	19.9192	0.7356	0.7048	0.7199	0.9592	1.0	0.8156	0.0606	0.0	0.7185	0.6855	0.6418	0.8226	0.8036	0.7641	0.6416	0.8152	0.5830	0.6648	0.5576	0.8025
467.1807	4.0	784	18.3366	0.7491	0.7535	0.7513	0.9624	1.0	0.8608	0.6491	0.0	0.7304	0.7740	0.6737	0.8094	0.8145	0.8034	0.7044	0.8202	0.5980	0.6999	0.6328	0.8221
467.1807	5.0	980	18.2944	0.7686	0.7503	0.7593	0.9629	0.6667	0.8875	0.7458	0.0	0.7472	0.7321	0.7125	0.8274	0.8453	0.8007	0.6836	0.8327	0.6541	0.7058	0.6227	0.8205
53.9369	6.0	1176	18.9329	0.7585	0.7894	0.7736	0.9648	0.6667	0.7907	0.6833	0.0	0.7707	0.7665	0.7120	0.8384	0.8476	0.8157	0.6921	0.8371	0.6336	0.7322	0.6695	0.8690
53.9369	7.0	1372	19.5750	0.7704	0.7822	0.7763	0.9650	0.5	0.8287	0.7	0.0	0.7796	0.7485	0.7356	0.8130	0.8397	0.8107	0.7262	0.8349	0.6524	0.7426	0.6786	0.8161
20.5433	8.0	1568	21.3628	0.7546	0.8024	0.7778	0.9641	0.6667	0.8588	0.7287	0.0	0.7911	0.7765	0.6806	0.8156	0.8496	0.8152	0.7453	0.8285	0.6667	0.7269	0.6627	0.8136
20.5433	9.0	1764	22.3521	0.7415	0.8176	0.7777	0.9641	0.5	0.8304	0.6963	0.0	0.8113	0.7553	0.6927	0.8254	0.8470	0.7913	0.7244	0.8449	0.6528	0.7524	0.6684	0.8295
20.5433	10.0	1960	22.8445	0.7467	0.8057	0.7751	0.9641	0.4	0.8488	0.736	0.0	0.8021	0.7712	0.6825	0.8156	0.8485	0.7956	0.7211	0.8409	0.6508	0.7339	0.6658	0.8353
9.7332	11.0	2156	23.1898	0.7552	0.8109	0.7821	0.9649	0.4	0.8652	0.7438	0.0	0.8028	0.7825	0.6862	0.8149	0.8556	0.8177	0.7235	0.8374	0.6721	0.7401	0.6690	0.8795
9.7332	12.0	2352	24.5698	0.7510	0.8150	0.7817	0.9648	0.4	0.8353	0.7317	0.0	0.8040	0.7925	0.6918	0.8137	0.8582	0.8115	0.7303	0.8303	0.6510	0.7598	0.6615	0.8727
5.2110	13.0	2548	24.8242	0.7747	0.7978	0.7861	0.9660	0.6667	0.8284	0.7059	0.0	0.8094	0.7827	0.7086	0.8082	0.8520	0.8198	0.7372	0.8418	0.6630	0.7506	0.6799	0.8623
5.2110	14.0	2744	25.8318	0.7623	0.8054	0.7833	0.9655	0.6667	0.8353	0.6777	0.0	0.8038	0.7760	0.7059	0.7989	0.8582	0.8172	0.7279	0.8405	0.6368	0.7573	0.6708	0.8623
5.2110	15.0	2940	26.0250	0.7676	0.8088	0.7877	0.9662	0.6667	0.8383	0.7154	0.0	0.8106	0.7816	0.6877	0.8057	0.8595	0.8218	0.7259	0.8404	0.6452	0.7594	0.6845	0.8623

Framework versions

• Transformers 5.8.0
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.22.2