wav2vec2_xls_r_300m_DigitalUmuganda_Afrivoice_Shona_20hr_v2

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: 0.6454
• Wer: 0.4589
• Cer: 0.0942

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
9.1544	0.9132	200	3.6919	1.0	1.0
3.1318	1.8265	400	2.9289	1.0	1.0
2.5747	2.7397	600	2.6079	1.0	0.9108
0.6358	3.6530	800	3.0760	1.0	0.9662
0.3581	4.5662	1000	1.7062	0.9991	0.6222
0.278	5.4795	1200	2.2421	0.9999	0.7555
0.2371	6.3927	1400	1.1771	0.9832	0.4230
0.209	7.3059	1600	1.7248	0.9987	0.6265
0.175	8.2192	1800	1.1461	0.9769	0.3771
0.1629	9.1324	2000	0.9195	0.9554	0.3226
0.1438	10.0457	2200	1.1127	0.9837	0.3919
0.1241	10.9589	2400	0.8292	0.9306	0.2766
0.1138	11.8721	2600	0.7719	0.8899	0.2485
0.0988	12.7854	2800	0.6755	0.8452	0.2052
0.0886	13.6986	3000	0.6168	0.8024	0.1815
0.0818	14.6119	3200	0.6716	0.8237	0.2015
0.0759	15.5251	3400	0.5018	0.7002	0.1375
0.0647	16.4384	3600	0.5972	0.7773	0.1681
0.062	17.3516	3800	0.5742	0.7406	0.1566
0.057	18.2648	4000	0.5820	0.7318	0.1630
0.0505	19.1781	4200	0.4585	0.5858	0.1123
0.0502	20.0913	4400	0.4341	0.5562	0.1008
0.0432	21.0046	4600	0.4241	0.5169	0.0967
0.0412	21.9178	4800	0.4216	0.5103	0.0923
0.0403	22.8311	5000	0.4430	0.5358	0.0970
0.0383	23.7443	5200	0.4680	0.5673	0.1046
0.0364	24.6575	5400	0.4472	0.5212	0.0981
0.0359	25.5708	5600	0.4221	0.5049	0.0886
0.0333	26.4840	5800	0.4129	0.4750	0.0835
0.0324	27.3973	6000	0.4257	0.4626	0.0834
0.0325	28.3105	6200	0.4004	0.4302	0.0754
0.0303	29.2237	6400	0.4347	0.5023	0.0864
0.0289	30.1370	6600	0.4113	0.4563	0.0824
0.029	31.0502	6800	0.4012	0.4568	0.0803
0.028	31.9635	7000	0.4144	0.4533	0.0826
0.0277	32.8767	7200	0.4071	0.4092	0.0722
0.0259	33.7900	7400	0.4101	0.4172	0.0731
0.0266	34.7032	7600	0.4150	0.4117	0.0730
0.0251	35.6164	7800	0.4122	0.4316	0.0769
0.0242	36.5297	8000	0.4113	0.4175	0.0747
0.0235	37.4429	8200	0.4046	0.4034	0.0711
0.0234	38.3562	8400	0.4193	0.4032	0.0699
0.0225	39.2694	8600	0.3984	0.3859	0.0671
0.0228	40.1826	8800	0.4267	0.3774	0.0657
0.0214	41.0959	9000	0.4046	0.3840	0.0664
0.0203	42.0091	9200	0.4198	0.3849	0.0670
0.0209	42.9224	9400	0.4260	0.3637	0.0620
0.0209	43.8356	9600	0.4096	0.3753	0.0661
0.0207	44.7489	9800	0.4170	0.3930	0.0671
0.0198	45.6621	10000	0.4158	0.4124	0.0691
0.0192	46.5753	10200	0.4133	0.3594	0.0625
0.0188	47.4886	10400	0.4111	0.3790	0.0653
0.0179	48.4018	10600	0.4144	0.3817	0.0657
0.018	49.3151	10800	0.4229	0.3782	0.0656
0.0172	50.2283	11000	0.4300	0.3807	0.0654
0.0181	51.1416	11200	0.4177	0.3601	0.0617
0.0171	52.0548	11400	0.4198	0.3841	0.0654
0.0176	52.9680	11600	0.4308	0.3675	0.0624
0.0163	53.8813	11800	0.4330	0.3647	0.0624
0.0164	54.7945	12000	0.4332	0.3832	0.0666
0.015	55.7078	12200	0.4433	0.3626	0.0628

Framework versions

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