wav2vec2_xls_r_300m_DigitalUmuganda_Afrivoice_Shona_10hr_v5

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: 0.5899
• Wer: 0.6249
• Cer: 0.1342

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
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 500
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
9.3891	1.8100	200	3.8736	1.0	1.0
3.2028	3.6199	400	2.9371	1.0	1.0
2.4728	5.4299	600	1.5567	1.0	0.4447
0.5836	7.2398	800	1.6129	0.9967	0.5418
0.3128	9.0498	1000	1.5253	0.9964	0.5396
0.2287	10.8597	1200	1.2166	0.9759	0.4185
0.1809	12.6697	1400	1.4519	0.9869	0.4705
0.1498	14.4796	1600	1.9091	0.9970	0.5663
0.1291	16.2896	1800	2.6398	0.9998	0.7305
0.1068	18.0995	2000	0.3955	0.5353	0.1006
0.0903	19.9095	2200	0.5005	0.6603	0.1384
0.0776	21.7195	2400	0.4863	0.6456	0.1223
0.0716	23.5294	2600	0.4941	0.6332	0.1245
0.061	25.3394	2800	0.5654	0.7182	0.1518
0.0557	27.1493	3000	0.5254	0.6574	0.1355
0.0513	28.9593	3200	0.5567	0.6706	0.1429
0.0463	30.7692	3400	0.5668	0.6532	0.1403
0.0426	32.5792	3600	0.5244	0.6183	0.1270
0.0393	34.3891	3800	0.5296	0.6264	0.1295
0.038	36.1991	4000	0.5641	0.6784	0.1443

Framework versions

• Transformers 4.44.0
• Pytorch 2.2.0+cu121
• Datasets 2.21.0
• Tokenizers 0.19.1