mms-1b-250_250h-hau-ft

This model is a fine-tuned version of facebook/mms-1b-all on the /MNT/MD0/SYNVOICES/DATA/HAUSA_250_250H - NA dataset. It achieves the following results on the evaluation set:

• Loss: 0.3308
• Wer: 0.3513
• Cer: 0.0887

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: 0.001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 2
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• total_eval_batch_size: 16
• 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: 2.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.3613	0.0320	500	0.4146	0.4071	0.1056
0.407	0.0641	1000	0.4004	0.4009	0.1034
0.342	0.0961	1500	0.3979	0.4061	0.1057
0.4212	0.1282	2000	0.3871	0.4145	0.1047
0.4009	0.1602	2500	0.3869	0.4030	0.1034
0.4283	0.1923	3000	0.3827	0.4032	0.1017
0.3129	0.2243	3500	0.3873	0.4055	0.1024
0.2488	0.2564	4000	0.3773	0.3883	0.0994
0.475	0.2884	4500	0.3820	0.3897	0.1000
0.4095	0.3205	5000	0.3688	0.3887	0.0990
0.3402	0.3525	5500	0.3722	0.3846	0.0982
0.2797	0.3846	6000	0.3696	0.3943	0.0987
0.4146	0.4166	6500	0.3725	0.3949	0.0993
0.3188	0.4487	7000	0.3661	0.3849	0.0987
0.2401	0.4807	7500	0.3653	0.3821	0.0977
0.2492	0.5128	8000	0.3594	0.3776	0.0964
0.3665	0.5448	8500	0.3685	0.3844	0.0982
0.4324	0.5769	9000	0.3651	0.3756	0.0957
0.2253	0.6089	9500	0.3625	0.3847	0.0970
0.2664	0.6410	10000	0.3580	0.3738	0.0947
0.4003	0.6730	10500	0.3556	0.3689	0.0936
0.3316	0.7051	11000	0.3570	0.3763	0.0952
0.2544	0.7371	11500	0.3531	0.3720	0.0941
0.2287	0.7692	12000	0.3579	0.3931	0.0972
0.3805	0.8012	12500	0.3558	0.3760	0.0957
0.3921	0.8333	13000	0.3533	0.3916	0.0975
0.3213	0.8653	13500	0.3541	0.3836	0.0957
0.4347	0.8973	14000	0.3528	0.3807	0.0956
0.3049	0.9294	14500	0.3548	0.3765	0.0951
0.3651	0.9614	15000	0.3506	0.3733	0.0942
0.2681	0.9935	15500	0.3512	0.3715	0.0936
0.2743	1.0255	16000	0.3489	0.3689	0.0932
0.3168	1.0576	16500	0.3495	0.3718	0.0937
0.2572	1.0896	17000	0.3476	0.3702	0.0931
0.3063	1.1217	17500	0.3398	0.3619	0.0916
0.2725	1.1537	18000	0.3471	0.3663	0.0923
0.3038	1.1858	18500	0.3448	0.3648	0.0920
0.1957	1.2178	19000	0.3453	0.3596	0.0906
0.2435	1.2498	19500	0.3429	0.3581	0.0908
0.2705	1.2819	20000	0.3400	0.3689	0.0921
0.2804	1.3139	20500	0.3414	0.3605	0.0915
0.2005	1.3460	21000	0.3417	0.3663	0.0915
0.3056	1.3780	21500	0.3427	0.3671	0.0917
0.2708	1.4101	22000	0.3393	0.3583	0.0909
0.2275	1.4421	22500	0.3396	0.3566	0.0902
0.3978	1.4742	23000	0.3383	0.3562	0.0902
0.39	1.5062	23500	0.3409	0.3548	0.0906
0.2801	1.5383	24000	0.3367	0.3604	0.0909
0.4354	1.5703	24500	0.3372	0.3565	0.0901
0.2681	1.6024	25000	0.3362	0.3549	0.0901
0.3148	1.6344	25500	0.3366	0.3542	0.0898
0.2573	1.6665	26000	0.3338	0.3518	0.0895
0.3137	1.6985	26500	0.3329	0.3551	0.0897
0.3578	1.7306	27000	0.3351	0.3510	0.0887
0.2697	1.7626	27500	0.3343	0.3549	0.0891
0.236	1.7947	28000	0.3330	0.3514	0.0887
0.1999	1.8267	28500	0.3326	0.3518	0.0890
0.3329	1.8588	29000	0.3305	0.3475	0.0883
0.2248	1.8908	29500	0.3310	0.3516	0.0889
0.2766	1.9229	30000	0.3308	0.3503	0.0886
0.4033	1.9549	30500	0.3311	0.3534	0.0890
0.3129	1.9870	31000	0.3308	0.3516	0.0887

Framework versions

• Transformers 4.48.1
• Pytorch 2.5.1+cu121
• Datasets 3.2.0
• Tokenizers 0.21.0