whisper-medium-basque

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

• Loss: 0.1445
• Wer: 8.2615

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: 96
• eval_batch_size: 48
• 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
0.1909	0.12	500	0.2820	22.5178
0.1253	0.25	1000	0.2133	16.1484
0.1046	0.37	1500	0.1899	12.7139
0.0874	0.5	2000	0.1793	11.4088
0.0836	0.62	2500	0.1621	10.5470
0.0726	0.74	3000	0.1597	10.1161
0.0707	0.87	3500	0.1498	9.4355
0.0652	0.99	4000	0.1470	8.5737
0.0416	1.11	4500	0.1482	8.5925
0.0415	1.24	5000	0.1490	8.6299
0.0394	1.36	5500	0.1474	8.0929
0.0381	1.49	6000	0.1425	8.3489
0.038	1.61	6500	0.1414	8.2990
0.0333	1.73	7000	0.1391	8.2553
0.0342	1.86	7500	0.1382	8.3864
0.0341	1.98	8000	0.1386	8.4301
0.0196	2.11	8500	0.1447	8.1429
0.0208	2.23	9000	0.1448	8.3115
0.018	2.35	9500	0.1449	8.3177
0.0172	2.48	10000	0.1445	8.2615

Framework versions

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