zuazo/whisper-large-v2-eu-from-es

Whisper Large-V2 Basque

This model is a fine-tuned version of zuazo/whisper-large-v2-es on the mozilla-foundation/common_voice_13_0 eu dataset. It achieves the following results on the evaluation set:

• Loss: 0.4245
• Wer: 11.8393

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: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 20000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0293	4.01	1000	0.2732	15.9484
0.0065	9.01	2000	0.3051	14.1136
0.0033	14.01	3000	0.3101	13.2407
0.0041	19.0	4000	0.3136	13.8300
0.0013	24.0	5000	0.3179	12.7364
0.0046	29.0	6000	0.3210	13.6640
0.0015	33.01	7000	0.3262	12.8093
0.0027	38.01	8000	0.3210	12.9612
0.0005	43.01	9000	0.3376	12.7850
0.0007	48.01	10000	0.3361	12.9126
0.0002	53.0	11000	0.3559	12.3739
0.0001	58.0	12000	0.3550	12.3355
0.0	63.0	13000	0.3852	12.1147
0.0	67.01	14000	0.3974	12.0134
0.0	72.01	15000	0.4072	11.9446
0.0	77.01	16000	0.4162	11.9203
0.0	82.01	17000	0.4245	11.8393
0.0	87.0	18000	0.4319	11.8616
0.0	92.0	19000	0.4375	11.8535
0.0	97.0	20000	0.4400	11.8656

Framework versions

• Transformers 4.33.0.dev0
• Pytorch 2.0.1+cu117
• Datasets 2.14.4
• Tokenizers 0.13.3