asr-africa/mms-300m_DigitalUmuganda_Afrivoice_Shona_1hr_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: 2.0090
• Wer: 0.9948
• Cer: 0.4189

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
68.8683	0.8696	5	16.0023	1.0	0.9538
67.5244	1.8261	10	15.5550	1.0	0.9430
61.8287	2.7826	15	12.7110	1.0	1.0
38.8336	3.9130	21	7.7816	1.0	1.0
28.1041	4.8696	26	5.3122	1.0	1.0
20.7205	5.8261	31	4.4709	1.0	1.0
17.9969	6.7826	36	4.1134	1.0	1.0
13.9168	7.9130	42	3.7606	1.0	1.0
15.5405	8.8696	47	3.6320	1.0	1.0
14.8928	9.8261	52	3.4072	1.0	1.0
13.9935	10.7826	57	3.2393	1.0	1.0
11.1442	11.9130	63	3.1078	1.0	1.0
12.9364	12.8696	68	3.0432	1.0	1.0
12.7554	13.8261	73	2.9975	1.0	1.0
12.5164	14.7826	78	2.9721	1.0	1.0
10.3626	15.9130	84	2.9607	1.0	1.0
12.3881	16.8696	89	2.9378	1.0	1.0
13.1535	17.8261	94	2.9394	1.0	1.0
12.2578	18.7826	99	2.9396	1.0	1.0
10.2163	19.9130	105	2.9266	1.0	1.0
12.2213	20.8696	110	2.9475	1.0	1.0
12.2561	21.8261	115	2.9099	1.0	1.0
12.2262	22.7826	120	2.9104	1.0	1.0
10.1776	23.9130	126	2.9074	1.0	1.0
12.157	24.8696	131	2.8994	1.0	1.0
12.1486	25.8261	136	2.9076	1.0	1.0
12.156	26.7826	141	2.9197	1.0	1.0
10.1287	27.9130	147	2.8963	1.0	1.0
12.1832	28.8696	152	2.9037	1.0	1.0
12.1224	29.8261	157	2.8914	1.0	1.0
12.094	30.7826	162	2.8870	1.0	1.0
10.0635	31.9130	168	2.9047	1.0	1.0
12.1146	32.8696	173	2.8729	1.0	1.0
12.0052	33.8261	178	2.8594	1.0	1.0
12.0124	34.7826	183	2.8509	1.0	1.0
10.0458	35.9130	189	2.9509	1.0	1.0
12.1021	36.8696	194	2.8455	1.0	1.0
11.8997	37.8261	199	2.8262	1.0	1.0
11.8208	38.7826	204	2.8231	1.0	1.0
9.8093	39.9130	210	2.8053	1.0	1.0
11.724	40.8696	215	2.7914	1.0	1.0
11.6571	41.8261	220	2.7760	1.0	1.0
11.5854	42.7826	225	2.7794	1.0	1.0
9.59	43.9130	231	2.7130	1.0	1.0
11.3268	44.8696	236	2.6681	1.0	0.9903
11.1653	45.8261	241	2.6518	1.0	0.9273
10.9978	46.7826	246	2.5816	1.0	0.9112
8.9489	47.9130	252	2.5036	1.0	0.9026
10.4282	48.8696	257	2.4504	1.0	0.8652
10.2606	49.8261	262	2.3633	1.0	0.8550
9.8352	50.7826	267	2.2941	1.0	0.8328
7.8901	51.9130	273	2.1868	1.0	0.8017
9.1182	52.8696	278	2.1071	1.0	0.7764
8.7192	53.8261	283	2.0319	1.0	0.6983
8.287	54.7826	288	1.9401	1.0	0.6283
6.5955	55.9130	294	1.8406	1.0	0.6137
7.5011	56.8696	299	1.7786	1.0	0.5562
7.1636	57.8261	304	1.7299	1.0	0.5691
6.8314	58.7826	309	1.6673	1.0	0.5061
5.4578	59.9130	315	1.6018	1.0	0.4657
6.2479	60.8696	320	1.5645	1.0	0.4674
5.8711	61.8261	325	1.5283	1.0	0.4500
5.6188	62.7826	330	1.4788	1.0	0.4256
4.4687	63.9130	336	1.4583	1.0	0.4233
5.0484	64.8696	341	1.4361	1.0	0.4096
4.8114	65.8261	346	1.4144	1.0	0.4062
4.4987	66.7826	351	1.4102	1.0	0.4007
3.5595	67.9130	357	1.4017	0.9998	0.3897
4.0309	68.8696	362	1.4033	0.9990	0.3850
3.8135	69.8261	367	1.3971	0.9981	0.3836
3.5936	70.7826	372	1.4172	0.9946	0.3778
2.8143	71.9130	378	1.4172	0.9803	0.3630
3.2126	72.8696	383	1.4275	0.9895	0.3667
3.0545	73.8261	388	1.4452	0.9735	0.3605
2.8559	74.7826	393	1.4479	0.9796	0.3606
2.2344	75.9130	399	1.4668	0.9679	0.3530
2.5623	76.8696	404	1.4827	0.9725	0.3532
2.4213	77.8261	409	1.5082	0.9803	0.3557
2.3092	78.7826	414	1.5162	0.9788	0.3519
1.8793	79.9130	420	1.5187	0.9672	0.3479
2.1443	80.8696	425	1.5260	0.9708	0.3518
2.0246	81.8261	430	1.5579	1.0002	0.3492
1.9691	82.7826	435	1.5867	1.0066	0.3532
1.6321	83.9130	441	1.5419	0.9769	0.3463
1.9106	84.8696	446	1.5800	0.9757	0.3477
1.8555	85.8261	451	1.5746	0.9669	0.3447
1.8178	86.7826	456	1.6088	0.9745	0.3452
1.4789	87.9130	462	1.5913	0.9803	0.3466
1.7375	88.8696	467	1.5916	0.9740	0.3454
1.6749	89.8261	472	1.6099	0.9647	0.3469
1.7118	90.7826	477	1.6322	1.0005	0.3522
1.3896	91.9130	483	1.6341	0.9715	0.3473
1.5778	92.8696	488	1.6404	0.9861	0.3469
1.607	93.8261	493	1.6349	0.9747	0.3475
1.6138	94.7826	498	1.6354	0.9740	0.3470
0.9601	95.2174	500	1.6371	0.9747	0.3467

Framework versions

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