wav2vec2_xls_r_300m_DigitalUmuganda_Afrivoice_Shona_50hr_v1

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: 0.3698
• Wer: 0.3938
• Cer: 0.0785

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• 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
9.205	0.1832	200	3.6211	1.0	1.0
3.1436	0.3663	400	2.9405	1.0	1.0
2.8535	0.5495	600	2.7629	1.0	0.9086
0.9706	0.7326	800	2.2864	1.0	0.8127
0.4797	0.9158	1000	1.8810	0.9999	0.7236
0.3773	1.0989	1200	1.4982	0.9987	0.6202
0.3256	1.2821	1400	1.3982	0.9981	0.5665
0.2972	1.4652	1600	1.0015	0.9896	0.4103
0.2913	1.6484	1800	0.8134	0.9836	0.3450
0.2792	1.8315	2000	0.4965	0.8095	0.1868
0.2701	2.0147	2200	0.4560	0.8199	0.1718
0.2314	2.1978	2400	0.3556	0.6607	0.1108
0.223	2.3810	2600	0.3025	0.5262	0.0872
0.2119	2.5641	2800	0.2845	0.4968	0.0781
0.2197	2.7473	3000	0.2595	0.4313	0.0679
0.2164	2.9304	3200	0.2635	0.4804	0.0724
0.1953	3.1136	3400	0.2598	0.4284	0.0689
0.1801	3.2967	3600	0.2723	0.4893	0.0775
0.1769	3.4799	3800	0.2504	0.3996	0.0652
0.1872	3.6630	4000	0.2477	0.3981	0.0655
0.1933	3.8462	4200	0.2608	0.4524	0.0717
0.185	4.0293	4400	0.2327	0.3352	0.0572
0.162	4.2125	4600	0.2352	0.3511	0.0580
0.1597	4.3956	4800	0.2326	0.3472	0.0576
0.1648	4.5788	5000	0.2303	0.3252	0.0543
0.1615	4.7619	5200	0.2326	0.3513	0.0574
0.1551	4.9451	5400	0.2256	0.3309	0.0553
0.1375	5.1282	5600	0.2290	0.3382	0.0546
0.1414	5.3114	5800	0.2389	0.3579	0.0607
0.1309	5.4945	6000	0.2290	0.3204	0.0551
0.143	5.6777	6200	0.2281	0.3042	0.0517
0.1443	5.8608	6400	0.2301	0.3287	0.0564
0.1405	6.0440	6600	0.2367	0.3406	0.0565
0.1107	6.2271	6800	0.2332	0.3096	0.0523
0.1193	6.4103	7000	0.2387	0.3065	0.0527
0.1217	6.5934	7200	0.2413	0.3361	0.0554
0.1184	6.7766	7400	0.2351	0.3138	0.0539
0.1264	6.9597	7600	0.2368	0.3360	0.0543
0.1068	7.1429	7800	0.2402	0.3232	0.0527
0.1036	7.3260	8000	0.2483	0.3294	0.0554
0.0995	7.5092	8200	0.2401	0.3201	0.0532

Framework versions

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