Whisper_Large_Mongolian

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

• eval_loss: 0.3309
• eval_wer: 29.1925
• eval_runtime: 1202.8614
• eval_samples_per_second: 1.539
• eval_steps_per_second: 0.193
• epoch: 5.64
• step: 3000

Training procedure

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 4000
• mixed_precision_training: Native AMP

Framework versions

• Transformers 4.39.3
• Pytorch 2.2.0
• Datasets 2.18.0
• Tokenizers 0.15.2