wav2vec2-xls-r-1b-cv8

This model is a fine-tuned version of facebook/wav2vec2-xls-r-1b on the MOZILLA-FOUNDATION/COMMON_VOICE_8_0 - PT dataset. It achieves the following results on the evaluation set:

Loss: 0.2007
Wer: 0.1838

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: 7.5e-05
train_batch_size: 4
eval_batch_size: 4
seed: 42
gradient_accumulation_steps: 4
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: 2000
num_epochs: 30.0
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
2.1172	0.32	500	1.2852	0.9783
1.4152	0.64	1000	0.6434	0.6105
1.4342	0.96	1500	0.4844	0.3989
1.4657	1.29	2000	0.5080	0.4490
1.4961	1.61	2500	0.4764	0.4264
1.4515	1.93	3000	0.4519	0.4068
1.3924	2.25	3500	0.4472	0.4132
1.4524	2.57	4000	0.4455	0.3939
1.4328	2.89	4500	0.4369	0.4069
1.3456	3.22	5000	0.4234	0.3774
1.3725	3.54	5500	0.4387	0.3789
1.3812	3.86	6000	0.4298	0.3825
1.3282	4.18	6500	0.4025	0.3703
1.3326	4.5	7000	0.3917	0.3502
1.3028	4.82	7500	0.3889	0.3582
1.293	5.14	8000	0.3859	0.3496
1.321	5.47	8500	0.3875	0.3576
1.3165	5.79	9000	0.3927	0.3589
1.2701	6.11	9500	0.4058	0.3621
1.2718	6.43	10000	0.4211	0.3916
1.2683	6.75	10500	0.3968	0.3620
1.2643	7.07	11000	0.4128	0.3848
1.2485	7.4	11500	0.3849	0.3727
1.2608	7.72	12000	0.3770	0.3474
1.2388	8.04	12500	0.3774	0.3574
1.2524	8.36	13000	0.3789	0.3550
1.2458	8.68	13500	0.3770	0.3410
1.2505	9.0	14000	0.3638	0.3403
1.2254	9.32	14500	0.3770	0.3509
1.2459	9.65	15000	0.3592	0.3349
1.2049	9.97	15500	0.3600	0.3428
1.2097	10.29	16000	0.3626	0.3347
1.1988	10.61	16500	0.3740	0.3269
1.1671	10.93	17000	0.3548	0.3245
1.1532	11.25	17500	0.3394	0.3140
1.1459	11.58	18000	0.3349	0.3156
1.1511	11.9	18500	0.3272	0.3110
1.1465	12.22	19000	0.3348	0.3084
1.1426	12.54	19500	0.3193	0.3027
1.1278	12.86	20000	0.3318	0.3021
1.149	13.18	20500	0.3169	0.2947
1.114	13.5	21000	0.3224	0.2986
1.1249	13.83	21500	0.3227	0.2921
1.0968	14.15	22000	0.3033	0.2878
1.0851	14.47	22500	0.2996	0.2863
1.0985	14.79	23000	0.3011	0.2843
1.0808	15.11	23500	0.2932	0.2759
1.069	15.43	24000	0.2919	0.2750
1.0602	15.76	24500	0.2959	0.2713
1.0369	16.08	25000	0.2931	0.2754
1.0573	16.4	25500	0.2920	0.2722
1.051	16.72	26000	0.2855	0.2632
1.0279	17.04	26500	0.2850	0.2649
1.0496	17.36	27000	0.2817	0.2585
1.0516	17.68	27500	0.2961	0.2635
1.0244	18.01	28000	0.2781	0.2589
1.0099	18.33	28500	0.2783	0.2565
1.0016	18.65	29000	0.2719	0.2537
1.0157	18.97	29500	0.2621	0.2449
0.9572	19.29	30000	0.2582	0.2427
0.9802	19.61	30500	0.2707	0.2468
0.9577	19.94	31000	0.2563	0.2389
0.9562	20.26	31500	0.2592	0.2382
0.962	20.58	32000	0.2539	0.2341
0.9541	20.9	32500	0.2505	0.2288
0.9587	21.22	33000	0.2486	0.2302
0.9146	21.54	33500	0.2461	0.2269
0.9215	21.86	34000	0.2387	0.2228
0.9105	22.19	34500	0.2405	0.2222
0.8949	22.51	35000	0.2316	0.2191
0.9153	22.83	35500	0.2358	0.2180
0.8907	23.15	36000	0.2369	0.2168
0.8973	23.47	36500	0.2323	0.2120
0.8878	23.79	37000	0.2293	0.2104
0.8818	24.12	37500	0.2302	0.2132
0.8919	24.44	38000	0.2262	0.2083
0.8473	24.76	38500	0.2257	0.2040
0.8516	25.08	39000	0.2246	0.2031
0.8451	25.4	39500	0.2198	0.2000
0.8288	25.72	40000	0.2199	0.1990
0.8465	26.05	40500	0.2165	0.1972
0.8305	26.37	41000	0.2128	0.1957
0.8202	26.69	41500	0.2127	0.1937
0.8223	27.01	42000	0.2100	0.1934
0.8322	27.33	42500	0.2076	0.1905
0.8139	27.65	43000	0.2054	0.1880
0.8299	27.97	43500	0.2026	0.1868
0.7937	28.3	44000	0.2045	0.1872
0.7972	28.62	44500	0.2025	0.1861
0.809	28.94	45000	0.2026	0.1858
0.813	29.26	45500	0.2013	0.1838
0.7718	29.58	46000	0.2010	0.1837
0.7929	29.9	46500	0.2008	0.1840

Framework versions

Transformers 4.17.0.dev0
Pytorch 1.10.2+cu102
Datasets 1.18.3.dev0
Tokenizers 0.11.0

lgris/wav2vec2-xls-r-1b-cv8

wav2vec2-xls-r-1b-cv8

Model description

Intended uses & limitations

Training and evaluation data

Training procedure

Training hyperparameters

Training results

Framework versions

Seed this model