wav2vec2_common_voice_accents_indian

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

• Loss: 0.2692

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: 48
• eval_batch_size: 4
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 384
• total_eval_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
4.5186	1.28	400	0.6937
0.3485	2.56	800	0.2323
0.2229	3.83	1200	0.2195
0.1877	5.11	1600	0.2147
0.1618	6.39	2000	0.2058
0.1434	7.67	2400	0.2077
0.132	8.95	2800	0.1995
0.1223	10.22	3200	0.2146
0.1153	11.5	3600	0.2117
0.1061	12.78	4000	0.2071
0.1003	14.06	4400	0.2219
0.0949	15.34	4800	0.2204
0.0889	16.61	5200	0.2162
0.0824	17.89	5600	0.2243
0.0784	19.17	6000	0.2323
0.0702	20.45	6400	0.2325
0.0665	21.73	6800	0.2334
0.0626	23.0	7200	0.2411
0.058	24.28	7600	0.2473
0.054	25.56	8000	0.2591
0.0506	26.84	8400	0.2577
0.0484	28.12	8800	0.2633
0.0453	29.39	9200	0.2692

Framework versions

• Transformers 4.17.0
• Pytorch 1.10.2+cu102
• Datasets 1.18.4
• Tokenizers 0.11.6