wav2vec2-large-xls-r-300m-tr-colab

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: 0.4121
• Wer: 0.3112

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
• distributed_type: multi-GPU
• 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	Wer
4.1868	1.83	400	0.9812	0.8398
0.691	3.67	800	0.5571	0.6298
0.3555	5.5	1200	0.4676	0.4779
0.2451	7.34	1600	0.4572	0.4541
0.1844	9.17	2000	0.4743	0.4389
0.1541	11.01	2400	0.4583	0.4300
0.1277	12.84	2800	0.4565	0.3950
0.1122	14.68	3200	0.4761	0.4087
0.0975	16.51	3600	0.4654	0.3786
0.0861	18.35	4000	0.4503	0.3667
0.0775	20.18	4400	0.4600	0.3581
0.0666	22.02	4800	0.4350	0.3504
0.0627	23.85	5200	0.4211	0.3349
0.0558	25.69	5600	0.4390	0.3333
0.0459	27.52	6000	0.4218	0.3185
0.0439	29.36	6400	0.4121	0.3112

Framework versions

• Transformers 4.11.3
• Pytorch 1.10.2+cu102
• Datasets 1.18.3
• Tokenizers 0.10.3