flima/openai-whisper-large-v3-fullFT-es_ecu911_martin_win30s

Whisper openai-whisper-large-v3

This model is a fine-tuned version of openai/whisper-large-v3 on the llamadas ecu911 dataset. It achieves the following results on the evaluation set:

• Loss: 0.2920
• Wer: 110.7983

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: 2e-05
• train_batch_size: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: constant
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.5302	0.9993	378	0.3469	38.3939
0.2016	1.9987	756	0.2801	60.4543
0.1027	2.9980	1134	0.2752	55.1432
0.0542	4.0	1513	0.2670	93.7130
0.032	4.9993	1891	0.2619	87.6820
0.0196	5.9987	2269	0.2810	56.8069
0.0158	6.9980	2647	0.2645	101.6477
0.0125	8.0	3026	0.2896	79.0274
0.0122	8.9993	3404	0.2679	82.7708
0.0097	9.9934	3780	0.2920	110.7983

Framework versions

• Transformers 4.45.2
• Pytorch 2.4.1+cu121
• Datasets 3.0.1
• Tokenizers 0.20.1