asr-africa/mms-300m_DigitalUmuganda_Afrivoice_Shona_5hr_v1

facebook/mms-300m

This model is a fine-tuned version of facebook/mms-300m on the DigitalUmuganda dataset. It achieves the following results on the evaluation set:

• Loss: 0.9857
• Wer: 0.5175
• Cer: 0.1156

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• 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: 100
• num_epochs: 100
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
54.8231	0.9818	27	5.5912	1.0	1.0
16.0974	1.9727	54	3.2813	1.0	1.0
12.3829	2.9636	81	2.9582	1.0	1.0
11.6369	3.9909	109	2.9327	1.0	1.0
11.78	4.9818	136	3.0289	1.0	1.0
11.7294	5.9727	163	2.8970	1.0	1.0
11.5829	6.9636	190	2.8451	1.0	1.0
10.9847	7.9909	218	2.7870	1.0	1.0
11.269	8.9818	245	2.6834	1.0	0.9652
10.1931	9.9727	272	2.2750	1.0	0.8276
8.6755	10.9636	299	1.8566	1.0	0.5798
6.7633	11.9909	327	1.4861	1.0	0.4707
5.7016	12.9818	354	1.2004	0.9976	0.3762
4.6139	13.9727	381	0.9838	0.9509	0.2874
3.7333	14.9636	408	0.8431	0.9037	0.2456
2.9457	15.9909	436	0.6929	0.8081	0.1956
2.4083	16.9818	463	0.5662	0.7055	0.1549
1.8317	17.9727	490	0.5390	0.6683	0.1435
1.4594	18.9636	517	0.4966	0.6362	0.1308
1.1558	19.9909	545	0.4624	0.5875	0.1182
0.9968	20.9818	572	0.4782	0.5759	0.1137
0.8777	21.9727	599	0.4626	0.5409	0.1046
0.7342	22.9636	626	0.4618	0.5367	0.1057
0.6756	23.9909	654	0.4537	0.5375	0.1032
0.5935	24.9818	681	0.4574	0.5195	0.0998
0.5394	25.9727	708	0.4524	0.5122	0.0974
0.4708	26.9636	735	0.4890	0.5034	0.0951
0.4191	27.9909	763	0.4569	0.5058	0.0949
0.4185	28.9818	790	0.5009	0.4964	0.0949
0.4048	29.9727	817	0.5135	0.5163	0.0983
0.3577	30.9636	844	0.4956	0.4942	0.0929
0.3449	31.9909	872	0.4670	0.4757	0.0919
0.3383	32.9818	899	0.4793	0.4903	0.0910
0.3013	33.9727	926	0.5074	0.4920	0.0907
0.2725	34.9636	953	0.5002	0.4842	0.0890
0.2641	35.9909	981	0.5312	0.4723	0.0882
0.2538	36.9818	1008	0.4744	0.4737	0.0863
0.2392	37.9727	1035	0.5041	0.4621	0.0863
0.2282	38.9636	1062	0.5037	0.4511	0.0848
0.2088	39.9909	1090	0.4988	0.4655	0.0869
0.2062	40.9818	1117	0.4873	0.4579	0.0842
0.204	41.9727	1144	0.4689	0.4521	0.0836
0.1856	42.9636	1171	0.5070	0.4555	0.0823
0.1847	43.9909	1199	0.5058	0.4458	0.0826
0.1899	44.9818	1226	0.4997	0.4334	0.0808
0.1716	45.9727	1253	0.4995	0.4244	0.0796
0.1772	46.9636	1280	0.4993	0.4399	0.0812
0.1612	47.9909	1308	0.4982	0.4343	0.0800
0.1645	48.9818	1335	0.4861	0.4321	0.0799
0.1596	49.9727	1362	0.4963	0.4236	0.0788
0.1544	50.9636	1389	0.5150	0.4358	0.0795
0.1356	51.9909	1417	0.5069	0.4470	0.0808
0.1445	52.9818	1444	0.5112	0.4343	0.0790
0.1381	53.9727	1471	0.5201	0.4146	0.0760
0.1355	54.9636	1498	0.4991	0.4110	0.0756
0.1309	55.9909	1526	0.5260	0.4397	0.0798
0.1359	56.9818	1553	0.5096	0.4285	0.0789
0.1181	57.9727	1580	0.5013	0.4224	0.0767
0.1227	58.9636	1607	0.5219	0.4125	0.0758
0.1127	59.9909	1635	0.5043	0.4210	0.0759
0.1083	60.9818	1662	0.4853	0.4010	0.0744
0.1152	61.9727	1689	0.5032	0.4054	0.0740
0.1142	62.9636	1716	0.5048	0.4086	0.0745
0.0975	63.9909	1744	0.5218	0.3996	0.0723
0.105	64.9818	1771	0.5210	0.4112	0.0740
0.094	65.9727	1798	0.5418	0.4073	0.0727
0.0987	66.9636	1825	0.5166	0.4008	0.0721
0.0958	67.9909	1853	0.5008	0.4098	0.0722
0.0936	68.9818	1880	0.5419	0.3988	0.0719
0.0896	69.9727	1907	0.5570	0.4219	0.0747
0.0853	70.9636	1934	0.5534	0.4117	0.0740
0.0793	71.9909	1962	0.5557	0.4078	0.0726
0.0805	72.9818	1989	0.5368	0.4018	0.0717
0.0875	73.9727	2016	0.5476	0.4049	0.0741
0.076	74.9636	2043	0.5561	0.4066	0.0729
0.0703	75.9909	2071	0.5527	0.4052	0.0722
0.0707	76.9818	2098	0.5543	0.3959	0.0713
0.0665	77.9727	2125	0.5628	0.4003	0.0708
0.0677	78.9636	2152	0.5413	0.3957	0.0699
0.0638	79.9909	2180	0.5498	0.3988	0.0706
0.0652	80.9818	2207	0.5507	0.3930	0.0699
0.061	81.9727	2234	0.5259	0.3881	0.0682
0.06	82.9636	2261	0.5397	0.3896	0.0684
0.0564	83.9909	2289	0.5441	0.3842	0.0677
0.0635	84.9818	2316	0.5372	0.3840	0.0678
0.0514	85.9727	2343	0.5504	0.3816	0.0683
0.0467	86.9636	2370	0.5573	0.3774	0.0674
0.0485	87.9909	2398	0.5604	0.3811	0.0674
0.0519	88.9818	2425	0.5459	0.3733	0.0665
0.0514	89.9727	2452	0.5411	0.3799	0.0668
0.0475	90.9636	2479	0.5369	0.3772	0.0664
0.0434	91.9909	2507	0.5510	0.3850	0.0672
0.0488	92.9818	2534	0.5488	0.3774	0.0659
0.046	93.9727	2561	0.5443	0.3794	0.0663
0.0463	94.9636	2588	0.5463	0.3806	0.0666
0.0399	95.9909	2616	0.5500	0.3796	0.0663
0.0401	96.9818	2643	0.5494	0.3769	0.0657
0.0431	97.9727	2670	0.5516	0.375	0.0657
0.0404	98.9636	2697	0.5523	0.3765	0.0659
0.0409	99.0818	2700	0.5523	0.3765	0.0661

Framework versions

• Transformers 4.47.0.dev0
• Pytorch 2.1.0+cu118
• Datasets 3.0.2
• Tokenizers 0.20.1