mmarron14/whisper-medium-cv17-es-5000-steps_proc3-def3

Whisper Medium CV17 Es 5000 steps with the same training configuration and processing of text as 500-steps_proc3-def3 -with filtering 30sec, customised optimizer, and processing of text using method 3- ; but with the values for some training_args scaled in proportion to 5000 - María Marrón

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: 0.1402
• Wer Ortho: 9.1690
• Wer: 5.1498

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: 2
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 32
• total_train_batch_size: 64
• optimizer: Use OptimizerNames.ADAMW_TORCH with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Wer Ortho
0.1806	0.5	1000	0.1703	6.2760	10.5261
0.1607	1.0	2000	0.1536	5.4669	9.6349
0.1599	0.6	3000	0.1525	5.3907	9.4987
0.1468	0.8	4000	0.1448	5.1333	9.2110
0.1508	1.0	5000	0.1402	5.1498	9.1690

Framework versions

• Transformers 4.53.2
• Pytorch 2.9.1+cu128
• Datasets 2.14.4
• Tokenizers 0.21.4