nicolarsen/whisper-medium-3-F

Whisper da-nst

This model is a fine-tuned version of openai/whisper-medium on the common_voice_17_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.8780
• Wer: 28.6353

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: 1e-05
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 11000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0096	4.01	1000	0.7403	31.2960
0.0046	9.0	2000	0.7646	29.8505
0.0016	13.02	3000	0.7695	30.8398
0.0009	18.01	4000	0.7821	31.2102
0.0006	22.02	5000	0.8035	31.6303
0.0011	27.01	6000	0.8169	29.6336
0.0001	32.0	7000	0.8244	29.6246
0.0	36.01	8000	0.8461	28.8205
0.0	41.01	9000	0.8633	28.7754
0.0	45.02	10000	0.8738	28.6986
0.0	50.01	11000	0.8780	28.6353

Framework versions

• Transformers 4.37.2
• Pytorch 2.2.0+cu121
• Datasets 2.18.0
• Tokenizers 0.15.1