wav2vec2_common_voice_accents_scotland

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: 0.2752

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
4.7171	1.28	400	1.1618
0.4391	2.56	800	0.2422
0.2259	3.83	1200	0.2071
0.1813	5.11	1600	0.2126
0.1531	6.39	2000	0.2010
0.1383	7.67	2400	0.2004
0.13	8.95	2800	0.2069
0.1193	10.22	3200	0.2081
0.1124	11.5	3600	0.2051
0.1023	12.78	4000	0.2175
0.097	14.06	4400	0.2261
0.0863	15.34	4800	0.2301
0.0823	16.61	5200	0.2334
0.079	17.89	5600	0.2252
0.0743	19.17	6000	0.2393
0.0696	20.45	6400	0.2481
0.0644	21.73	6800	0.2416
0.064	23.0	7200	0.2449
0.0584	24.28	7600	0.2660
0.0544	25.56	8000	0.2630
0.0523	26.84	8400	0.2677
0.0494	28.12	8800	0.2730
0.0462	29.39	9200	0.2752

Framework versions

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