zuazo/whisper-medium-gl-train

Whisper Medium Galician

This model is a fine-tuned version of openai/whisper-medium on the mozilla-foundation/common_voice_13_0 gl dataset. It achieves the following results on the evaluation set:

• Loss: 0.2829
• Wer: 6.4331

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: 64
• eval_batch_size: 32
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0825	5.81	1000	0.1863	7.0141
0.0138	11.63	2000	0.2197	6.6873
0.0059	17.44	3000	0.2436	6.5922
0.0033	23.26	4000	0.2536	6.5040
0.002	29.07	5000	0.2606	6.6925
0.002	34.88	6000	0.2690	6.6890
0.001	40.7	7000	0.2734	6.5853
0.0007	46.51	8000	0.2821	6.4798
0.0006	52.33	9000	0.2829	6.4331
0.0007	58.14	10000	0.2847	6.4452

Framework versions

• Transformers 4.37.2
• Pytorch 2.2.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1