whisper-tiny-basque

This model is a fine-tuned version of openai/whisper-tiny on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.4296
• Wer: 28.7020

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
0.7245	0.21	500	0.9431	64.8287
0.4752	0.42	1000	0.6923	48.8196
0.4115	0.63	1500	0.6084	40.8976
0.3696	0.84	2000	0.5597	37.1048
0.3336	1.05	2500	0.5289	36.0029
0.3189	1.26	3000	0.5082	33.5204
0.3007	1.47	3500	0.4903	32.4969
0.2935	1.68	4000	0.4802	31.6632
0.2839	1.89	4500	0.4695	30.7243
0.2594	2.1	5000	0.4615	30.2538
0.2507	2.31	5500	0.4547	29.4160
0.2574	2.52	6000	0.4487	29.5852
0.2523	2.73	6500	0.4429	28.5473
0.2471	2.94	7000	0.4393	29.3562
0.2329	3.15	7500	0.4373	29.2592
0.2346	3.36	8000	0.4343	28.5947
0.2322	3.57	8500	0.4332	29.0136
0.2304	3.78	9000	0.4305	28.1387
0.2299	3.99	9500	0.4300	28.0768
0.236	4.2	10000	0.4296	28.7020

Framework versions

• Transformers 4.36.0
• Pytorch 2.1.1+cu121
• Datasets 2.8.0
• Tokenizers 0.15.2