wav2vec2_common_voice_accents_3

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: 0.0042

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
4.584	1.27	400	1.1439
0.481	2.55	800	0.1986
0.2384	3.82	1200	0.1060
0.1872	5.1	1600	0.1016
0.158	6.37	2000	0.0942
0.1427	7.64	2400	0.0646
0.1306	8.92	2800	0.0612
0.1197	10.19	3200	0.0423
0.1129	11.46	3600	0.0381
0.1054	12.74	4000	0.0326
0.0964	14.01	4400	0.0293
0.0871	15.29	4800	0.0239
0.0816	16.56	5200	0.0168
0.0763	17.83	5600	0.0202
0.0704	19.11	6000	0.0224
0.0669	20.38	6400	0.0208
0.063	21.66	6800	0.0074
0.0585	22.93	7200	0.0126
0.0548	24.2	7600	0.0086
0.0512	25.48	8000	0.0080
0.0487	26.75	8400	0.0052
0.0455	28.03	8800	0.0062
0.0433	29.3	9200	0.0042

Framework versions

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