whisper-base-basque

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

• Loss: 0.2499
• Wer: 58.5175

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: 192
• eval_batch_size: 96
• 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
0.4078	0.25	500	0.5613	137.5859
0.2533	0.5	1000	0.3971	103.7717
0.1994	0.74	1500	0.3350	72.6240
0.1723	0.99	2000	0.3100	54.8270
0.1403	1.24	2500	0.2895	47.9955
0.1318	1.49	3000	0.2799	63.6193
0.1279	1.73	3500	0.2711	76.5205
0.1192	1.98	4000	0.2666	59.5729
0.104	2.23	4500	0.2604	54.1401
0.0986	2.48	5000	0.2601	53.6406
0.0929	2.73	5500	0.2540	59.4667
0.0971	2.97	6000	0.2522	41.1265
0.0806	3.22	6500	0.2526	52.0420
0.0812	3.47	7000	0.2508	54.2213
0.0816	3.72	7500	0.2498	55.6263
0.0799	3.96	8000	0.2511	62.8513
0.0723	4.21	8500	0.2500	55.0456
0.0724	4.46	9000	0.2498	59.0483
0.0707	4.71	9500	0.2502	53.7030
0.0685	4.96	10000	0.2499	58.5175

Framework versions

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