wav2vec2-xlsr-luganda-radio-mh

This model is a fine-tuned version of sulaimank/wav2vec2-xlsr-luganda on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4395
• Wer: 0.4305
• Cer: 0.1208

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: 1e-05
• train_batch_size: 8
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• 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
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.8346	1.0	1234	0.5410	0.4906	0.1462
0.7095	2.0	2468	0.5064	0.4830	0.1402
0.6739	3.0	3702	0.4869	0.4789	0.1376
0.6475	4.0	4936	0.4764	0.4650	0.1339
0.6288	5.0	6170	0.4673	0.4614	0.1314
0.6147	6.0	7404	0.4631	0.4529	0.1291
0.5997	7.0	8638	0.4597	0.4462	0.1276
0.5884	8.0	9872	0.4563	0.4475	0.1265
0.5775	9.0	11106	0.4488	0.4404	0.1242
0.5652	10.0	12340	0.4467	0.4395	0.1235
0.5639	11.0	13574	0.4441	0.4336	0.1228
0.5515	12.0	14808	0.4427	0.4372	0.1231
0.5478	13.0	16042	0.4427	0.4350	0.1222
0.5434	14.0	17276	0.4442	0.4323	0.1216
0.5355	15.0	18510	0.4402	0.4336	0.1218
0.5341	16.0	19744	0.4411	0.4327	0.1216
0.5289	17.0	20978	0.4402	0.4323	0.1216
0.527	18.0	22212	0.4401	0.4323	0.1212
0.5271	19.0	23446	0.4393	0.4291	0.1204
0.5253	20.0	24680	0.4395	0.4305	0.1208

Framework versions

• Transformers 4.57.1
• Pytorch 2.9.1+cu128
• Datasets 3.6.0
• Tokenizers 0.22.1