output

This model is a fine-tuned version of cahya/wav2vec2-base-turkish-artificial-cv on the COMMON_VOICE - TR dataset. It achieves the following results on the evaluation set:

• Loss: 0.1822
• Wer: 0.1423

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: 7.5e-07
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 2000
• num_epochs: 1.0

Training results

Framework versions

• Transformers 4.17.0.dev0
• Pytorch 1.10.1+cu102
• Datasets 1.18.2
• Tokenizers 0.10.3