whisper-small-dv-full

This model is a fine-tuned version of openai/whisper-small on the arrow dataset. It achieves the following results on the evaluation set:

• eval_loss: 0.0273
• eval_wer_ortho: 15.8343
• eval_wer: 2.3726
• eval_runtime: 11624.8972
• eval_samples_per_second: 3.774
• eval_steps_per_second: 0.079
• epoch: 0.6569
• step: 4000

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• distributed_type: multi-GPU
• num_devices: 3
• total_train_batch_size: 48
• total_eval_batch_size: 48
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: constant_with_warmup
• lr_scheduler_warmup_steps: 50
• training_steps: 4000
• mixed_precision_training: Native AMP

Framework versions

• Transformers 4.51.3
• Pytorch 2.6.0+cu124
• Datasets 3.5.0
• Tokenizers 0.21.1