wolbanking77-afro-xlmr-large

This model is a fine-tuned version of Davlan/afro-xlmr-large on the WolBanking77 dataset. It achieves the following results on the evaluation set:

• Loss: 2.2959
• F1: 0.5810

Model description

See WolBanking77: Wolof Banking Speech Intent Classification Dataset paper

Intended uses & limitations

• Customer Intent Detection
• Machine Translation in French & Wolof
• Automatic Speech Recognition in Wolof
• Comparing different machine learning models for Intent Classification

Training and evaluation data

See WolBanking77 dataset.

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 2e-05
• train_batch_size: 4
• 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
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	F1
4.2932	1.0	1958	4.0790	0.0315
2.9364	2.0	3916	2.7736	0.3511
2.288	3.0	5874	2.3849	0.4881
1.992	4.0	7832	2.2629	0.5396
1.6931	5.0	9790	2.2066	0.5694
1.3725	6.0	11748	2.2224	0.5770
1.2085	7.0	13706	2.2959	0.5810
0.9111	8.0	15664	2.4032	0.5704
0.8359	9.0	17622	2.4982	0.5719
0.6609	10.0	19580	2.6449	0.5720
0.4818	11.0	21538	2.7611	0.5691
0.417	12.0	23496	2.9075	0.5677
0.285	13.0	25454	3.0922	0.5570
0.1873	14.0	27412	3.2292	0.5582
0.1321	15.0	29370	3.3421	0.5646
0.0938	16.0	31328	3.4509	0.5694
0.0704	17.0	33286	3.5159	0.5695
0.0423	18.0	35244	3.5586	0.5740
0.0262	19.0	37202	3.6021	0.5652
0.0228	20.0	39160	3.6177	0.5654

Framework versions

• Transformers 4.57.1
• Pytorch 2.8.0+cu128
• Datasets 4.2.0
• Tokenizers 0.22.1