wav2vec2-large-xls-r-300m-spanish-small-v3

This model is a fine-tuned version of jhonparra18/wav2vec2-large-xls-r-300m-spanish-custom on the common_voice dataset. It achieves the following results on the evaluation set:

• Loss: 0.3986
• Wer: 0.1980

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: 0.0004
• 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: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 25
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2372	0.26	400	0.3011	0.2660
0.3413	0.53	800	0.3559	0.3228
0.3598	0.79	1200	0.3753	0.3400
0.3529	1.05	1600	0.3385	0.2979
0.3133	1.32	2000	0.3559	0.3056
0.3158	1.58	2400	0.3364	0.2994
0.3092	1.85	2800	0.3210	0.2876
0.2919	2.11	3200	0.3460	0.3010
0.2666	2.37	3600	0.3543	0.3036
0.2819	2.64	4000	0.3477	0.2959
0.283	2.9	4400	0.3492	0.2968
0.2484	3.16	4800	0.3647	0.2993
0.2371	3.43	5200	0.3601	0.2942
0.2382	3.69	5600	0.3656	0.3019
0.2425	3.96	6000	0.3379	0.2873
0.2092	4.22	6400	0.3385	0.2736
0.2171	4.48	6800	0.3503	0.2889
0.2185	4.75	7200	0.3289	0.2727
0.2236	5.01	7600	0.3447	0.2771
0.1882	5.27	8000	0.3586	0.2860
0.1986	5.54	8400	0.3404	0.2829
0.2055	5.8	8800	0.3561	0.2869
0.196	6.06	9200	0.3633	0.2811
0.1748	6.33	9600	0.3703	0.2818
0.1758	6.59	10000	0.3525	0.2816
0.1819	6.86	10400	0.3581	0.2765
0.1715	7.12	10800	0.3480	0.2628
0.1606	7.38	11200	0.3490	0.2703
0.1632	7.65	11600	0.3461	0.2706
0.1638	7.91	12000	0.3458	0.2673
0.1552	8.17	12400	0.3646	0.2732
0.154	8.44	12800	0.3706	0.2726
0.1512	8.7	13200	0.3609	0.2683
0.149	8.97	13600	0.3610	0.2668
0.1357	9.23	14000	0.3693	0.2740
0.1375	9.49	14400	0.3677	0.2625
0.1391	9.76	14800	0.3795	0.2762
0.1378	10.02	15200	0.3541	0.2592
0.1197	10.28	15600	0.3562	0.2507
0.1259	10.55	16000	0.3612	0.2584
0.1266	10.81	16400	0.3470	0.2527
0.1199	11.07	16800	0.3721	0.2571
0.1157	11.34	17200	0.3734	0.2571
0.1107	11.6	17600	0.3730	0.2589
0.1148	11.87	18000	0.3648	0.2536
0.1095	12.13	18400	0.3746	0.2521
0.1047	12.39	18800	0.3566	0.2530
0.1043	12.66	19200	0.3794	0.2545
0.1066	12.92	19600	0.3548	0.2439
0.0974	13.18	20000	0.3702	0.2461
0.0978	13.45	20400	0.3721	0.2492
0.095	13.71	20800	0.3599	0.2467
0.0963	13.97	21200	0.3650	0.2402
0.0902	14.24	21600	0.3689	0.2459
0.0898	14.5	22000	0.3832	0.2452
0.0865	14.77	22400	0.3982	0.2436
0.0911	15.03	22800	0.3785	0.2398
0.0793	15.29	23200	0.3731	0.2396
0.0806	15.56	23600	0.3626	0.2372
0.0789	15.82	24000	0.3707	0.2356
0.0779	16.08	24400	0.3850	0.2368
0.078	16.35	24800	0.3831	0.2363
0.0732	16.61	25200	0.3947	0.2287
0.0733	16.88	25600	0.3928	0.2374
0.0721	17.14	26000	0.3943	0.2324
0.0676	17.4	26400	0.3793	0.2311
0.0682	17.67	26800	0.3958	0.2257
0.0714	17.93	27200	0.3890	0.2322
0.0673	18.19	27600	0.3872	0.2229
0.0613	18.46	28000	0.3828	0.2226
0.0621	18.72	28400	0.3812	0.2214
0.0622	18.98	28800	0.3919	0.2212
0.0576	19.25	29200	0.4000	0.2205
0.0581	19.51	29600	0.3953	0.2203
0.0573	19.78	30000	0.3947	0.2190
0.0576	20.04	30400	0.3909	0.2156
0.0551	20.3	30800	0.4178	0.2153
0.0525	20.57	31200	0.3935	0.2152
0.0522	20.83	31600	0.4054	0.2151
0.0519	21.09	32000	0.3877	0.2135
0.0479	21.36	32400	0.4119	0.2107
0.0472	21.62	32800	0.3967	0.2091
0.048	21.89	33200	0.3812	0.2057
0.0458	22.15	33600	0.3931	0.2043
0.0459	22.41	34000	0.3937	0.2049
0.0448	22.68	34400	0.3900	0.2056
0.0432	22.94	34800	0.4050	0.2049
0.0425	23.2	35200	0.3985	0.2014
0.0415	23.47	35600	0.3976	0.2013
0.0403	23.73	36000	0.4031	0.2018
0.04	23.99	36400	0.3996	0.2000
0.039	24.26	36800	0.3977	0.1993
0.0406	24.52	37200	0.3967	0.2000
0.0391	24.79	37600	0.3986	0.1980

Framework versions

• Transformers 4.17.0.dev0
• Pytorch 1.10.2+cu102
• Datasets 1.18.2.dev0
• Tokenizers 0.11.0