nb-bert-base-user-needs

This model is a fine-tuned version of NbAiLab/nb-bert-base 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: 1.0600
• Accuracy: 0.8479
• F1: 0.8319
• Precision: 0.8315
• Recall: 0.8479

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: 16
• 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	98	1.1222	0.6263	0.5185	0.5076	0.6263
No log	2.0	196	1.0066	0.7216	0.6436	0.5899	0.7216
No log	3.0	294	0.8540	0.7577	0.7037	0.6760	0.7577
No log	4.0	392	0.8621	0.7603	0.6998	0.6568	0.7603
No log	5.0	490	0.8062	0.7887	0.7500	0.7449	0.7887
0.91	6.0	588	0.7465	0.8041	0.7660	0.7636	0.8041
0.91	7.0	686	0.6324	0.8247	0.8163	0.8187	0.8247
0.91	8.0	784	0.7333	0.7964	0.7703	0.7740	0.7964
0.91	9.0	882	0.6590	0.8325	0.8208	0.8106	0.8325
0.91	10.0	980	0.9854	0.8196	0.7890	0.7920	0.8196
0.4246	11.0	1078	0.7023	0.8247	0.8054	0.8138	0.8247
0.4246	12.0	1176	0.8995	0.8325	0.8120	0.8068	0.8325
0.4246	13.0	1274	0.8589	0.8299	0.8145	0.8058	0.8299
0.4246	14.0	1372	0.9859	0.8376	0.8151	0.8123	0.8376
0.4246	15.0	1470	0.8452	0.8402	0.8318	0.8341	0.8402
0.1637	16.0	1568	1.1156	0.8351	0.8157	0.8196	0.8351
0.1637	17.0	1666	1.1514	0.8325	0.8122	0.8218	0.8325
0.1637	18.0	1764	1.0092	0.8428	0.8266	0.8320	0.8428
0.1637	19.0	1862	1.0368	0.8351	0.8229	0.8287	0.8351
0.1637	20.0	1960	1.0600	0.8479	0.8319	0.8315	0.8479
0.0391	21.0	2058	1.1046	0.8428	0.8293	0.8269	0.8428
0.0391	22.0	2156	1.1178	0.8454	0.8262	0.8280	0.8454
0.0391	23.0	2254	1.1103	0.8428	0.8268	0.8295	0.8428
0.0391	24.0	2352	1.1179	0.8428	0.8274	0.8313	0.8428
0.0391	25.0	2450	1.1134	0.8402	0.8233	0.8254	0.8402

Framework versions

• Transformers 4.17.0
• Pytorch 1.10.2+cu113
• Datasets 2.3.2
• Tokenizers 0.12.1