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: 0.3235
• Wer: 22.3867

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
0.5126	0.33	500	0.6979	53.3533
0.3117	0.66	1000	0.5084	37.0488
0.2552	0.99	1500	0.4231	30.1736
0.2102	1.32	2000	0.3911	29.4055
0.1925	1.65	2500	0.3738	26.7079
0.1767	1.98	3000	0.3588	24.9469
0.1572	2.31	3500	0.3541	25.3466
0.1503	2.64	4000	0.3432	24.1726
0.1485	2.97	4500	0.3347	22.8987
0.131	3.3	5000	0.3336	23.1173
0.1294	3.63	5500	0.3305	23.2422
0.1287	3.96	6000	0.3292	22.6240
0.1153	4.29	6500	0.3270	22.5990
0.1176	4.62	7000	0.3243	23.8167
0.1153	4.95	7500	0.3262	22.7051
0.1093	5.28	8000	0.3248	23.8479
0.1129	5.61	8500	0.3244	23.7480
0.107	5.94	9000	0.3233	22.4054
0.1029	6.27	9500	0.3236	22.7801
0.1056	6.61	10000	0.3235	22.3867

Framework versions

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