wav2vec2-large-xls-r-53m-gl-jupyter5

This model is a fine-tuned version of facebook/wav2vec2-large-xlsr-53 on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.1025
• Wer: 0.0625

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: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_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: 60
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.6862	3.36	400	0.2455	0.2344
0.1517	6.72	800	0.1195	0.1233
0.0772	10.08	1200	0.1219	0.1155
0.0472	13.44	1600	0.1162	0.1034
0.0357	16.8	2000	0.1070	0.1006
0.0307	20.17	2400	0.1131	0.1013
0.0258	23.53	2800	0.1163	0.0847
0.0229	26.89	3200	0.1100	0.0858
0.0183	30.25	3600	0.1062	0.0810
0.0182	33.61	4000	0.1068	0.0800
0.0151	36.97	4400	0.1088	0.0780
0.0138	40.33	4800	0.1062	0.0737
0.0121	43.69	5200	0.1061	0.0722
0.0088	47.06	5600	0.1055	0.0670
0.008	50.42	6000	0.1059	0.0646
0.007	53.78	6400	0.1020	0.0634
0.0065	57.14	6800	0.1025	0.0625

Framework versions

• Transformers 4.11.3
• Pytorch 1.9.1+cu111
• Datasets 1.18.3
• Tokenizers 0.10.3