Wav2vec2-afr

This model is a fine-tuned version of asr-africa/wav2vec2-xls-r-ewe-100-hours on the LEONEL-MAIA/EWE_DATASET_SPLITTED - DEFAULT dataset. It achieves the following results on the evaluation set:

• Loss: 0.1822
• Wer: 0.2055

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: 2
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 16
• 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_steps: 500
• num_epochs: 60.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2638	0.3017	500	0.3208	0.3275
0.1603	0.6034	1000	0.2497	0.2649
0.142	0.9050	1500	0.2145	0.2488
0.1645	1.2064	2000	0.2023	0.2252
0.1818	1.5080	2500	0.1984	0.2177
0.1827	1.8097	3000	0.1995	0.2204
0.1859	2.1110	3500	0.1994	0.2240
0.2488	2.4127	4000	0.2040	0.2250
0.2292	2.7144	4500	0.1991	0.2260
0.2489	3.0157	5000	0.1845	0.2137
0.4731	3.3174	5500	0.1854	0.2121
0.2497	3.6191	6000	0.1888	0.2102
0.2872	3.9207	6500	0.1915	0.2197
0.248	4.2220	7000	0.1920	0.2199
0.2048	4.5237	7500	0.1822	0.2055
0.1977	4.8254	8000	0.1850	0.2094
0.1459	5.1267	8500	0.1905	0.2153
0.1471	5.4284	9000	0.1864	0.2023
0.1528	5.7301	9500	0.1829	0.2089
0.0668	6.0314	10000	0.1908	0.2050

Framework versions

• Transformers 4.50.3
• Pytorch 2.7.0+cu126
• Datasets 3.5.0
• Tokenizers 0.21.1