zuazo/whisper-large-eu-cv16_1

Whisper Large Basque

This model is a fine-tuned version of openai/whisper-large on the mozilla-foundation/common_voice_16_1 eu dataset. It achieves the following results on the evaluation set:

• Loss: 0.4111
• Wer: 8.1444

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: 32
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 256
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 40000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.004	10.04	1000	0.2314	10.6603
0.0028	20.08	2000	0.2480	10.2783
0.0027	30.11	3000	0.2492	10.0379
0.0005	40.15	4000	0.2753	9.3784
0.0016	50.19	5000	0.2489	9.3003
0.0006	60.23	6000	0.2599	9.0023
0.0011	70.26	7000	0.2606	8.9378
0.0005	80.3	8000	0.2723	8.9270
0.0001	90.34	9000	0.2764	8.5304
0.0011	100.38	10000	0.2668	8.8977
0.0001	110.41	11000	0.2856	8.3701
0.0	120.45	12000	0.3045	8.2890
0.0	130.49	13000	0.3149	8.2441
0.0	140.53	14000	0.3241	8.2285
0.0	150.56	15000	0.3336	8.2060
0.0	160.6	16000	0.3433	8.1601
0.0	170.64	17000	0.3537	8.1806
0.0	180.68	18000	0.3634	8.1874
0.0	190.72	19000	0.3738	8.1786
0.0	200.75	20000	0.3848	8.2441
0.0	210.79	21000	0.3952	8.2324
0.0	220.83	22000	0.4030	8.2480
0.0001	230.87	23000	0.2919	8.4268
0.0	240.9	24000	0.3137	8.1865
0.0	250.94	25000	0.3271	8.1884
0.0	260.98	26000	0.3378	8.1825
0.0	271.02	27000	0.3472	8.1865
0.0	281.05	28000	0.3556	8.2031
0.0	291.09	29000	0.3637	8.2099
0.0	301.13	30000	0.3710	8.1933
0.0	311.17	31000	0.3781	8.1874
0.0	321.2	32000	0.3845	8.1679
0.0	331.24	33000	0.3905	8.1601
0.0	341.28	34000	0.3971	8.1640
0.0	351.32	35000	0.4022	8.1611
0.0	361.36	36000	0.4046	8.1562
0.0	371.39	37000	0.4073	8.1523
0.0	381.43	38000	0.4093	8.1493
0.0	391.47	39000	0.4107	8.1513
0.0	401.51	40000	0.4111	8.1444

Framework versions

• Transformers 4.37.2
• Pytorch 2.2.0+cu121
• Datasets 2.16.1
• Tokenizers 0.15.1