wav2vec2_common_voice_accents_4

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: 0.0047

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
4.615	1.28	400	0.8202
0.3778	2.56	800	0.1587
0.2229	3.85	1200	0.1027
0.1799	5.13	1600	0.0879
0.1617	6.41	2000	0.0772
0.1474	7.69	2400	0.0625
0.134	8.97	2800	0.0498
0.1213	10.26	3200	0.0429
0.1186	11.54	3600	0.0434
0.1118	12.82	4000	0.0312
0.1026	14.1	4400	0.0365
0.0951	15.38	4800	0.0321
0.0902	16.67	5200	0.0262
0.0843	17.95	5600	0.0208
0.0744	19.23	6000	0.0140
0.0718	20.51	6400	0.0204
0.0694	21.79	6800	0.0133
0.0636	23.08	7200	0.0104
0.0609	24.36	7600	0.0084
0.0559	25.64	8000	0.0050
0.0527	26.92	8400	0.0089
0.0495	28.21	8800	0.0058
0.0471	29.49	9200	0.0047

Framework versions

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