wav2vec2_common_voice_accents_indian_only_rerun

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: 1.2807

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: 588
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
4.6205	25.0	400	1.4584
0.3427	50.0	800	1.8377
0.1213	75.0	1200	1.6086
0.0643	100.0	1600	1.5136
0.0433	125.0	2000	1.4882
0.0323	150.0	2400	1.2204
0.0265	175.0	2800	1.3034
0.0206	200.0	3200	1.2866
0.0191	225.0	3600	1.2337
0.0148	250.0	4000	1.1729
0.0121	275.0	4400	1.2059
0.0105	300.0	4800	1.1246
0.01	325.0	5200	1.1397
0.0098	350.0	5600	1.1684
0.0073	375.0	6000	1.1030
0.0061	400.0	6400	1.2077
0.0049	425.0	6800	1.2653
0.0044	450.0	7200	1.1587
0.0037	475.0	7600	1.2283
0.0033	500.0	8000	1.1897
0.0026	525.0	8400	1.2633
0.0023	550.0	8800	1.2571
0.002	575.0	9200	1.2807

Framework versions

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