zuazo/whisper-large-v3-eu-cv22.0

openai/whisper-large-v3

This model is a fine-tuned version of openai/whisper-large-v3 on the common_voice_22_0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3653
• Wer: 7.4202

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: 3.75e-05
• train_batch_size: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• 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: 100000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.014	10.8234	5000	0.2178	9.1690
0.0066	21.6457	10000	0.2418	8.8613
0.003	32.4680	15000	0.2610	8.9864
0.0021	43.2904	20000	0.2620	8.7776
0.0028	54.1127	25000	0.2799	9.7522
0.002	64.9361	30000	0.2723	8.9146
0.0011	75.7584	35000	0.2742	8.6509
0.0019	86.5807	40000	0.2881	9.0929
0.0008	97.4030	45000	0.2848	8.2451
0.0003	108.2254	50000	0.2906	8.7016
0.0001	119.0477	55000	0.2925	8.6069
0.0012	129.8711	60000	0.2904	8.9611
0.0	140.6934	65000	0.3061	8.1082
0.0001	151.5157	70000	0.2946	8.3897
0.0	162.3380	75000	0.3021	8.5317
0.0	173.1603	80000	0.3179	8.0305
0.0	183.9837	85000	0.3386	7.7600
0.0	194.8061	90000	0.3542	7.5554
0.0	205.6284	95000	0.3627	7.4633
0.0	216.4507	100000	0.3653	7.4202

Framework versions

• Transformers 4.52.3
• Pytorch 2.6.0+cu124
• Datasets 3.6.0
• Tokenizers 0.21.1