VMadalina/whisper-medium-news-augmented2

Whisper Large Ro - VM3

This model is a fine-tuned version of openai/whisper-medium on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.2002
• Wer: 13.6187
• Cer: 5.2529

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: 5e-06
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 32
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_steps: 300
• training_steps: 8000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.1553	0.6031	1000	0.2150	18.6389	7.9857
0.0962	1.2063	2000	0.2028	17.7711	6.7850
0.0984	1.8094	3000	0.1927	15.0127	5.8714
0.0594	2.4125	4000	0.1942	14.1240	5.3363
0.0387	3.0157	5000	0.1937	12.9822	4.6948
0.0375	3.6188	6000	0.1976	13.5857	5.0185
0.0292	4.2220	7000	0.1997	13.5109	5.2041
0.03	4.8251	8000	0.2002	13.6187	5.2529

Framework versions

• Transformers 4.50.1
• Pytorch 2.6.0+cu124
• Datasets 2.19.1
• Tokenizers 0.21.1