custom_BERT_NER

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

• Loss: 0.207071
• Perf P: 0.829268
• Perf R: 0.944444
• Inst P: 0.933333
• Inst R: 0.875000
• Comp P: 0.962617
• Comp R: 0.865546
• Precision: 0.862745
• Recall: 0.846154
• F1: 0.854369
• Accuracy: 0.952260

Model description

This model is for identifying performers, instrumentation, and composers of the music played in the concert from a brief introduction of a concert.

Tags:
PERF: Performer(s)
INST: Instrumentation
COMP: Composer(s)
MUSIC: Music title(s)
PER: Other name(s)
OTH: Other instrument(s)
OTHP: Other music title(s)
ORG: Companies, festivals, orchetras, ensembles, etc.
LOC: Country names, halls, etc.
MISC: Other miscellaneous nouns, including competitions.

Training and evaluation data

This model is trained ane evaluated on a custome dataset: jamie613/custom_NER
The set contains 150 samples of concert introductions in Mandarine.
The dataset is divide into training set (135 samples) and evaluation set (15 samples).

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 1
• eval_batch_size: 1
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• num_epochs: 50
• metric_for_best_model = 'eval_f1'
• greater_is_better = True
• load_best_model_at_end = True
• early_stoping_patience = 3

Training results

Training Loss	Epoch	Step	Validation Loss	Perf P	Perf R	Inst P	Inst R	Comp P	Comp R	Precision	Recall	F1	Accuracy
0.8629	1.0	135	0.3555	0.6951	0.7917	0.5176	0.6875	0.8455	0.7815	0.6913	0.6095	0.6478	0.8848
0.2867	2.0	270	0.2387	0.6275	0.8889	0.7719	0.6875	0.93	0.7815	0.7778	0.7663	0.7720	0.9265
0.1715	3.0	405	0.1832	0.8193	0.9444	0.875	0.7656	0.8636	0.7983	0.8186	0.8077	0.8131	0.9446
0.1027	4.0	540	0.2056	0.875	0.875	0.75	0.7969	0.9630	0.8739	0.8254	0.8180	0.8217	0.9441
0.0707	5.0	675	0.2007	0.825	0.9167	0.9245	0.7656	0.9423	0.8235	0.8378	0.8328	0.8353	0.9468
0.0517	6.0	810	0.2402	0.8415	0.9583	0.8889	0.75	0.93	0.7815	0.8311	0.8225	0.8268	0.9403
0.0359	7.0	945	0.2071	0.8293	0.9444	0.9333	0.875	0.9626	0.8655	0.8627	0.8462	0.8544	0.9523
0.0269	8.0	1080	0.2171	0.8415	0.9583	0.9608	0.7656	0.9604	0.8151	0.8411	0.8299	0.8354	0.9486
0.0196	9.0	1215	0.2317	0.8718	0.9444	0.8788	0.9062	0.9558	0.9076	0.8505	0.8417	0.8461	0.9510
0.0126	10.0	1350	0.2578	0.8161	0.9861	0.8923	0.9062	0.9537	0.8655	0.8495	0.8432	0.8463	0.9470

Framework versions

• Transformers 4.40.0
• Pytorch 2.2.1+cu121
• Datasets 2.19.0
• Tokenizers 0.19.1