wav2vec2-large-xls-r-300m-common_voice-tr-ft

This model is a fine-tuned version of facebook/wav2vec2-xls-r-300m on the COMMON_VOICE - TR dataset. It achieves the following results on the evaluation set:

• Loss: 0.4179
• Wer: 0.3071
• Cer: 0.0736

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: 0.0005
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• num_devices: 8
• total_train_batch_size: 64
• total_eval_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 100.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.7638	9.09	500	0.4763	0.5313	0.1333
0.5739	18.18	1000	0.4007	0.4357	0.1099
0.4343	27.27	1500	0.3819	0.4060	0.1012
0.4401	36.36	2000	0.3991	0.3954	0.1001
0.2647	45.45	2500	0.3901	0.3689	0.0914
0.2656	54.55	3000	0.4284	0.3463	0.0852
0.2586	63.64	3500	0.4084	0.3297	0.0804
0.2041	72.73	4000	0.3907	0.3193	0.0781
0.4265	81.82	4500	0.4265	0.3120	0.0755
0.2041	90.91	5000	0.4240	0.3071	0.0736

Framework versions

• Transformers 4.13.0.dev0
• Pytorch 1.9.0+cu111
• Datasets 1.15.2.dev0
• Tokenizers 0.10.3