doku-berteus

This model is a fine-tuned version of ixa-ehu/berteus-base-cased on the basque_glue dataset. It achieves the following results on the evaluation set:

• Loss: 0.7077
• Micro F1: 0.7745
• Precision: 0.7745
• Recall: 0.7745
• F1 Class 0: 0.7290
• F1 Class 1: 0.2857
• F1 Class 2: 0.8256
• F1 Class 3: 0.7758
• F1 Class 4: 0.7534
• F1 Class 5: 0.8409
• F1 Class 6: 0.5378
• F1 Class 7: 0.375
• F1 Class 8: 0.8333
• F1 Class 9: 0.7769
• F1 Class 10: 0.8136
• F1 Class 11: 0.1538

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: 2e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 1
• 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
• num_epochs: 10

Training results

Training Loss	Epoch	Step	Validation Loss	Micro F1	Precision	Recall	F1 Class 0	F1 Class 1	F1 Class 2	F1 Class 3	F1 Class 4	F1 Class 5	F1 Class 6	F1 Class 7	F1 Class 8	F1 Class 9	F1 Class 10	F1 Class 11
0.8177	1.0	537	0.8337	0.7458	0.7458	0.7458	0.6906	0.0	0.8304	0.7453	0.6777	0.8086	0.4685	0.4828	0.8272	0.7709	0.7788	0.0
0.5226	2.0	1074	0.8253	0.7534	0.7534	0.7534	0.7152	0.08	0.8302	0.7599	0.6620	0.8278	0.5113	0.4706	0.8280	0.7826	0.7698	0.2727
0.3745	3.0	1611	0.9744	0.7458	0.7458	0.7458	0.7262	0.0714	0.7946	0.7526	0.6929	0.7967	0.5312	0.5574	0.8232	0.7449	0.7758	0.4615
0.1982	4.0	2148	1.1651	0.7518	0.7518	0.7518	0.7099	0.2069	0.8145	0.7632	0.6719	0.7911	0.5138	0.4898	0.8473	0.7560	0.7751	0.3704
0.0823	5.0	2685	1.4201	0.7421	0.7421	0.7421	0.6981	0.2286	0.8128	0.7428	0.6772	0.7884	0.5484	0.5185	0.8412	0.7692	0.7593	0.2609

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu121
• Datasets 3.1.0
• Tokenizers 0.21.0