arpagon/whisper-large-v2-es

Whisper Large v2 Spanish

This model is a fine-tuned version of openai/whisper-large-v2 on the mozilla-foundation/common_voice_11_0 es dataset. It achieves the following results on the evaluation set:

• Loss: 0.1702
• Wer google/fleurs: 4.89
• Wer mozilla-foundation/common_voice_11_0: 5.2882

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: 8
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• 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
0.1738	0.1	1000	0.2031	7.0384
0.2108	1.01	2000	0.1885	6.6668
0.1599	1.11	3000	0.1814	6.5342
0.0794	2.01	4000	0.1792	6.0314
0.0477	2.11	5000	0.1936	6.1795
0.0341	3.02	6000	0.2038	6.0113
0.0264	3.12	7000	0.2111	5.8410
0.0608	4.02	8000	0.1824	5.9067
0.0523	4.12	9000	0.1768	5.3941
0.0984	5.03	10000	0.1702	5.2882

Framework versions

• Transformers 4.26.0.dev0
• Pytorch 2.0.0.dev20221210+cu117
• Datasets 2.7.1.dev0
• Tokenizers 0.13.2