multilingual-google/mt5-base-lumasaba-ner-v1

This model is a fine-tuned version of google/mt5-base on the Beijuka/Multilingual_PII_NER_dataset dataset. It achieves the following results on the evaluation set:

• Loss: 0.7358
• Precision: 0.7910
• Recall: 0.7946
• F1: 0.7928
• Accuracy: 0.8100

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: 5e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Precision	Recall	F1	Accuracy
No log	1.0	398	2.7430	0.3351	0.45	0.3841	0.3355
4.9558	2.0	796	2.3019	0.2968	0.4930	0.3705	0.3133
2.7275	3.0	1194	2.0554	0.4145	0.4320	0.4231	0.3836
2.3526	4.0	1592	1.9726	0.7027	0.3305	0.4495	0.3887
2.3526	5.0	1990	1.6121	0.5668	0.5273	0.5463	0.5036
2.1054	6.0	2388	1.5167	0.4953	0.5711	0.5305	0.5334
1.8628	7.0	2786	1.3548	0.5812	0.6125	0.5964	0.5917
1.633	8.0	3184	1.2491	0.592	0.6359	0.6132	0.6126
1.4669	9.0	3582	1.1306	0.6485	0.6617	0.6551	0.6539
1.4669	10.0	3980	1.0705	0.6427	0.6773	0.6596	0.6722
1.2985	11.0	4378	0.8972	0.7033	0.7445	0.7233	0.7229
1.1477	12.0	4776	0.9050	0.7087	0.7336	0.7209	0.7220
1.0512	13.0	5174	0.8057	0.7515	0.7609	0.7562	0.7582
0.9269	14.0	5572	0.7670	0.7545	0.7805	0.7673	0.7718
0.9269	15.0	5970	0.6996	0.7830	0.8063	0.7945	0.7901
0.8489	16.0	6368	0.7031	0.7842	0.8063	0.7951	0.7935
0.7588	17.0	6766	0.6762	0.7891	0.8156	0.8022	0.8059
0.7432	18.0	7164	0.6648	0.7901	0.8234	0.8064	0.8080
0.6983	19.0	7562	0.6598	0.7991	0.8234	0.8111	0.8114
0.6983	20.0	7960	0.6626	0.7937	0.8203	0.8068	0.8097

Framework versions

• Transformers 4.55.4
• Pytorch 2.8.0+cu126
• Datasets 4.0.0
• Tokenizers 0.21.4