whisper-meduim-mongolian

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

• Loss: 0.3098
• Wer: 26.8664

Training hyperparameters

The following hyperparameters were used during training:

• learning_rate: 1e-05
• train_batch_size: 4
• 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: 2000
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.3034	0.9398	2000	0.4135	45.1152
0.1443	1.8797	4000	0.3127	35.3290
0.0618	2.8195	6000	0.3038	31.0534
0.0179	3.7594	8000	0.3042	28.3673
0.0028	4.6992	10000	0.3098	26.8664

Framework versions

• Transformers 4.40.1
• Pytorch 2.2.0
• Datasets 2.19.0
• Tokenizers 0.19.1