wav2vec2-large-xlsr-sw

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the common_voice_11_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4334
• Wer: 0.3231

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 30
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.3405	0.88	400	2.8882	1.0000
1.083	1.77	800	0.5223	0.5849
0.4721	2.65	1200	0.3921	0.4667
0.3793	3.54	1600	0.3725	0.4257
0.3264	4.42	2000	0.3646	0.4179
0.294	5.31	2400	0.3542	0.4104
0.2623	6.19	2800	0.3576	0.3892
0.2408	7.08	3200	0.3516	0.3876
0.2229	7.96	3600	0.3580	0.3877
0.206	8.85	4000	0.3466	0.3683
0.1991	9.73	4400	0.3306	0.3783
0.1863	10.62	4800	0.3605	0.3707
0.1743	11.5	5200	0.3483	0.3703
0.1678	12.39	5600	0.3645	0.3618
0.1547	13.27	6000	0.3671	0.3589
0.152	14.16	6400	0.3733	0.3568
0.144	15.04	6800	0.3684	0.3486
0.136	15.93	7200	0.3558	0.3493
0.1262	16.81	7600	0.3748	0.3486
0.1222	17.7	8000	0.3774	0.3466
0.1164	18.58	8400	0.3840	0.3427
0.1108	19.47	8800	0.3988	0.3438
0.1072	20.35	9200	0.4020	0.3384
0.1008	21.24	9600	0.4013	0.3375
0.0982	22.12	10000	0.4162	0.3361
0.0951	23.01	10400	0.4107	0.3346
0.0923	23.89	10800	0.4248	0.3337
0.0866	24.78	11200	0.4151	0.3295
0.0875	25.66	11600	0.4211	0.3310
0.0813	26.55	12000	0.4303	0.3290
0.0775	27.43	12400	0.4334	0.3249
0.0759	28.32	12800	0.4312	0.3240
0.0758	29.2	13200	0.4334	0.3231

Framework versions

• Transformers 4.37.0.dev0
• Pytorch 2.1.1+cu121
• Datasets 2.16.1
• Tokenizers 0.15.0