asr-africa/mms-300m_DigitalUmuganda_Afrivoice_Shona_10hr_v2

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: 0.8585
• Wer: 0.4104
• Cer: 0.0907

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
33.7728	0.9955	55	3.2848	1.0	1.0
12.0407	1.9910	110	2.9063	1.0	1.0
11.7013	2.9864	165	2.8895	1.0	1.0
11.2973	4.0	221	2.8169	1.0	1.0
11.1719	4.9955	276	2.5708	1.0	0.8738
8.8604	5.9910	331	1.6831	1.0	0.5577
6.0356	6.9864	386	1.1403	0.9983	0.3583
4.3072	8.0	442	0.8494	0.9471	0.2649
3.3072	8.9955	497	0.6382	0.8545	0.1923
2.3828	9.9910	552	0.4925	0.6959	0.1443
1.7086	10.9864	607	0.3803	0.5268	0.1041
1.2672	12.0	663	0.3669	0.4825	0.0934
1.0349	12.9955	718	0.3209	0.4346	0.0825
0.8498	13.9910	773	0.3147	0.4300	0.0818
0.7259	14.9864	828	0.3151	0.4054	0.0742
0.6193	16.0	884	0.3132	0.4200	0.0748
0.5381	16.9955	939	0.3256	0.3895	0.0711
0.4792	17.9910	994	0.3441	0.3863	0.0699
0.4266	18.9864	1049	0.3323	0.3855	0.0692
0.3717	20.0	1105	0.3396	0.3871	0.0671
0.3503	20.9955	1160	0.3394	0.3772	0.0679
0.3254	21.9910	1215	0.3383	0.3868	0.0682
0.2808	22.9864	1270	0.3784	0.3808	0.0684
0.2534	24.0	1326	0.3449	0.3629	0.0652
0.2467	24.9955	1381	0.3540	0.3725	0.0664
0.2271	25.9910	1436	0.3677	0.3512	0.0630
0.2197	26.9864	1491	0.3623	0.3664	0.0645
0.2041	28.0	1547	0.3791	0.3950	0.0653
0.2013	28.9955	1602	0.3656	0.3515	0.0619
0.1877	29.9910	1657	0.3700	0.3533	0.0632
0.1763	30.9864	1712	0.3907	0.3594	0.0638
0.1651	32.0	1768	0.3825	0.3713	0.0652
0.1697	32.9955	1823	0.3777	0.3546	0.0624
0.1699	33.9910	1878	0.4016	0.3536	0.0631
0.1553	34.9864	1933	0.4027	0.3547	0.0620
0.1438	36.0	1989	0.4027	0.3444	0.0612
0.1526	36.9955	2044	0.4131	0.3510	0.0613
0.1388	37.9910	2099	0.3921	0.3594	0.0617
0.1334	38.9864	2154	0.3944	0.3519	0.0614
0.1269	40.0	2210	0.4067	0.3486	0.0611
0.1332	40.9955	2265	0.3779	0.3416	0.0594
0.1239	41.9910	2320	0.4081	0.3432	0.0603
0.1172	42.9864	2375	0.3946	0.3356	0.0587
0.1174	44.0	2431	0.4208	0.3333	0.0582
0.1168	44.9955	2486	0.3811	0.3297	0.0585
0.1118	45.9910	2541	0.4092	0.3406	0.0596
0.1061	46.9864	2596	0.4050	0.3248	0.0573
0.1019	48.0	2652	0.4025	0.3250	0.0572
0.0959	48.9955	2707	0.4168	0.3306	0.0579
0.102	49.9910	2762	0.4353	0.3300	0.0571
0.0967	50.9864	2817	0.4308	0.3307	0.0587
0.0953	52.0	2873	0.4258	0.3227	0.0567
0.0936	52.9955	2928	0.4012	0.3262	0.0555
0.0917	53.9910	2983	0.4252	0.3219	0.0549
0.0851	54.9864	3038	0.4243	0.3331	0.0570
0.0829	56.0	3094	0.4353	0.3224	0.0558
0.0761	56.9955	3149	0.4437	0.3299	0.0569
0.074	57.9910	3204	0.4178	0.3231	0.0551
0.0731	58.9864	3259	0.4297	0.3249	0.0560
0.0683	60.0	3315	0.4271	0.3215	0.0549
0.0698	60.9955	3370	0.4462	0.3197	0.0539
0.0722	61.9910	3425	0.4369	0.3157	0.0535
0.0654	62.9864	3480	0.4259	0.3208	0.0545
0.0653	64.0	3536	0.4431	0.3169	0.0544
0.0636	64.9955	3591	0.4397	0.3149	0.0535
0.0619	65.9910	3646	0.4410	0.3195	0.0549
0.0622	66.9864	3701	0.4306	0.3199	0.0545
0.0558	68.0	3757	0.4444	0.3099	0.0534
0.0595	68.9955	3812	0.4487	0.3123	0.0530
0.0558	69.9910	3867	0.4535	0.3110	0.0534
0.0487	70.9864	3922	0.4498	0.3105	0.0530
0.0512	72.0	3978	0.4411	0.3073	0.0524
0.0496	72.9955	4033	0.4345	0.3135	0.0533
0.0505	73.9910	4088	0.4520	0.3080	0.0521
0.0431	74.9864	4143	0.4432	0.3044	0.0518
0.0456	76.0	4199	0.4548	0.2995	0.0514
0.045	76.9955	4254	0.4503	0.2997	0.0505
0.0443	77.9910	4309	0.4552	0.3042	0.0517
0.0435	78.9864	4364	0.4646	0.2986	0.0505
0.0407	80.0	4420	0.4461	0.3075	0.0509
0.0396	80.9955	4475	0.4507	0.3031	0.0510
0.0401	81.9910	4530	0.4454	0.3023	0.0508
0.0352	82.9864	4585	0.4473	0.2991	0.0503
0.0328	84.0	4641	0.4550	0.3067	0.0515
0.037	84.9955	4696	0.4513	0.2971	0.0502
0.0358	85.9910	4751	0.4447	0.2955	0.0497
0.034	86.9864	4806	0.4504	0.2952	0.0496
0.0313	88.0	4862	0.4674	0.2965	0.0500
0.0269	88.9955	4917	0.4620	0.2967	0.0496
0.0304	89.9910	4972	0.4541	0.2927	0.0490
0.029	90.9864	5027	0.4503	0.2958	0.0492
0.0293	92.0	5083	0.4555	0.2935	0.0492
0.027	92.9955	5138	0.4598	0.2955	0.0491
0.0285	93.9910	5193	0.4629	0.2946	0.0492
0.0279	94.9864	5248	0.4582	0.2933	0.0490
0.0264	96.0	5304	0.4576	0.2916	0.0490
0.0258	96.9955	5359	0.4605	0.2933	0.0491
0.0247	97.9910	5414	0.4591	0.2948	0.0491
0.0226	98.9864	5469	0.4610	0.2952	0.0492
0.0226	99.5475	5500	0.4610	0.2945	0.0491

Framework versions

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