whisper-tiny-basque

This model is a fine-tuned version of openai/whisper-tiny on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.3232
• Wer: 22.4241

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: 256
• eval_batch_size: 128
• 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
0.5114	0.33	500	0.6930	52.5228
0.3117	0.66	1000	0.5082	37.1487
0.2552	0.99	1500	0.4230	30.2610
0.2102	1.32	2000	0.3911	28.6062
0.1925	1.65	2500	0.3733	26.7453
0.1768	1.98	3000	0.3585	24.9657
0.1572	2.31	3500	0.3539	25.3466
0.1503	2.64	4000	0.3429	24.1913
0.1485	2.97	4500	0.3344	22.9050
0.131	3.3	5000	0.3335	22.9799
0.1294	3.63	5500	0.3303	23.1797
0.1287	3.96	6000	0.3290	23.0736
0.1153	4.29	6500	0.3267	22.0619
0.1176	4.62	7000	0.3241	24.1164
0.1153	4.95	7500	0.3260	23.9603
0.1093	5.28	8000	0.3247	23.8541
0.1129	5.61	8500	0.3242	22.8488
0.1069	5.94	9000	0.3231	22.4116
0.1029	6.27	9500	0.3233	22.5678
0.1056	6.61	10000	0.3232	22.4241

Framework versions

• Transformers 4.38.0
• Pytorch 2.1.1+cu121
• Datasets 2.8.0
• Tokenizers 0.15.2