wav2vec2-xls-r-asheshi-akan-10-hours

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.4559
• Wer: 0.0723
• Cer: 0.0459

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: 0.0003
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• 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
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 50

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
7.3442	2.7322	500	1.8242	0.9988	0.6319
0.8252	5.4645	1000	0.4331	0.2104	0.0993
0.3898	8.1967	1500	0.3435	0.1475	0.0839
0.2863	10.9290	2000	0.3124	0.1342	0.0782
0.2271	13.6612	2500	0.3027	0.1294	0.0670
0.1939	16.3934	3000	0.3051	0.1013	0.0580
0.1656	19.1257	3500	0.3653	0.0902	0.0527
0.1453	21.8579	4000	0.3289	0.0902	0.0515
0.1255	24.5902	4500	0.3513	0.0927	0.0566
0.1112	27.3224	5000	0.3732	0.0833	0.0515
0.0989	30.0546	5500	0.3910	0.0778	0.0461
0.0846	32.7869	6000	0.4495	0.0905	0.0598
0.0745	35.5191	6500	0.4240	0.0766	0.0479
0.064	38.2514	7000	0.4331	0.0812	0.0497
0.0584	40.9836	7500	0.4331	0.0756	0.0472
0.051	43.7158	8000	0.4533	0.0757	0.0459
0.0477	46.4481	8500	0.4495	0.0708	0.0450
0.0428	49.1803	9000	0.4559	0.0723	0.0459

Framework versions

• Transformers 4.48.3
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0