This model is a fine-tuned version of NbAiLab/nb-bert-large on a dataset of 2000 articles from Bergens Tidende, published between 06/01/2020 and 02/02/2020. These articles are labelled as one of six classes / user needs, as introduced by the BBC in 2017. It achieves the following results on the evaluation set:

Loss: 1.0102
Accuracy: 0.8900
F1: 0.8859
Precision: 0.8883
Recall: 0.8900

Model description

More information needed

Intended uses & limitations

More information needed

Training and evaluation data

More information needed

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

learning_rate: 3e-05
train_batch_size: 8
eval_batch_size: 64
seed: 42
optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
lr_scheduler_type: linear
lr_scheduler_warmup_steps: 500
num_epochs: 25
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Accuracy	F1	Precision	Recall
No log	1.0	195	0.6790	0.8082	0.7567	0.7679	0.8082
No log	2.0	390	0.5577	0.8465	0.8392	0.8364	0.8465
0.8651	3.0	585	0.5494	0.8338	0.8191	0.8145	0.8338
0.8651	4.0	780	0.5453	0.8517	0.8386	0.8293	0.8517
0.8651	5.0	975	0.8855	0.8491	0.8298	0.8444	0.8491
0.3707	6.0	1170	0.7282	0.8645	0.8526	0.8581	0.8645
0.3707	7.0	1365	0.8797	0.8619	0.8537	0.8573	0.8619
0.1092	8.0	1560	0.9120	0.8491	0.8520	0.8579	0.8491
0.1092	9.0	1755	1.0700	0.8696	0.8615	0.8669	0.8696
0.1092	10.0	1950	1.0599	0.8670	0.8654	0.8701	0.8670
0.0355	11.0	2145	1.0808	0.8670	0.8656	0.8685	0.8670
0.0355	12.0	2340	1.0102	0.8900	0.8859	0.8883	0.8900
0.0002	13.0	2535	1.0236	0.8849	0.8812	0.8824	0.8849
0.0002	14.0	2730	1.0358	0.8875	0.8833	0.8841	0.8875
0.0002	15.0	2925	1.0476	0.8875	0.8833	0.8841	0.8875
0.0001	16.0	3120	1.0559	0.8798	0.8764	0.8776	0.8798
0.0001	17.0	3315	1.0648	0.8798	0.8754	0.8765	0.8798
0.0001	18.0	3510	1.0720	0.8798	0.8754	0.8765	0.8798
0.0001	19.0	3705	1.0796	0.8824	0.8775	0.8783	0.8824
0.0001	20.0	3900	1.0862	0.8798	0.8739	0.8745	0.8798
0.0	21.0	4095	1.0917	0.8798	0.8739	0.8745	0.8798
0.0	22.0	4290	1.0973	0.8798	0.8739	0.8745	0.8798
0.0	23.0	4485	1.1007	0.8798	0.8739	0.8745	0.8798
0.0	24.0	4680	1.1029	0.8798	0.8739	0.8745	0.8798
0.0	25.0	4875	1.1037	0.8798	0.8739	0.8745	0.8798

Framework versions

Transformers 4.17.0
Pytorch 1.10.2+cu113
Datasets 2.3.2
Tokenizers 0.12.1

thusken/nb-bert-large-user-needs

nb-bert-large-user-needs

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

thusken/
nb-bert-large-user-needs