zuazo/whisper-base-eu-cv16_1

Whisper Base Basque

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

• Loss: 0.5038
• Wer: 16.1765

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: 2.5e-05
• train_batch_size: 128
• eval_batch_size: 64
• seed: 42
• gradient_accumulation_steps: 2
• 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
0.0225	10.0	1000	0.3059	19.0812
0.0037	20.0	2000	0.3530	18.4618
0.0012	30.0	3000	0.3724	17.9332
0.0004	40.0	4000	0.4025	17.8951
0.0002	50.0	5000	0.4245	17.8951
0.0001	60.0	6000	0.4459	17.9772
0.0001	70.0	7000	0.4665	18.0163
0.0	80.0	8000	0.4882	18.1081
0.0003	90.0	9000	0.3803	16.3807
0.0001	100.0	10000	0.4047	16.2293
0.0001	110.0	11000	0.4207	16.2420
0.0001	120.0	12000	0.4353	16.2879
0.0	130.0	13000	0.4502	16.3700
0.0	140.0	14000	0.4653	16.5087
0.0	150.0	15000	0.4805	16.4393
0.0	160.0	16000	0.4964	16.4941
0.0	170.0	17000	0.5128	16.5107
0.0	180.0	18000	0.5285	16.6377
0.0	190.0	19000	0.5457	16.6572
0.0102	200.0	20000	0.4229	18.1902
0.0	210.0	21000	0.4498	16.2117
0.0	220.0	22000	0.4646	16.2146
0.0	230.0	23000	0.4754	16.1961
0.0	240.0	24000	0.4853	16.1863
0.0	250.0	25000	0.4946	16.1912
0.0	260.0	26000	0.5038	16.1765
0.0	270.0	27000	0.5133	16.2215
0.0	280.0	28000	0.5228	16.2224
0.0	290.0	29000	0.5326	16.2557
0.0	300.0	30000	0.5427	16.2420
0.0	310.0	31000	0.5525	16.2635
0.0	320.0	32000	0.5624	16.2957
0.0	330.0	33000	0.5706	16.3299
0.0	340.0	34000	0.5798	16.3534
0.0	350.0	35000	0.5880	16.3495
0.0	360.0	36000	0.5948	16.3622
0.0	370.0	37000	0.6005	16.3934
0.0	380.0	38000	0.6045	16.3876
0.0	390.0	39000	0.6074	16.4325
0.0	400.0	40000	0.6085	16.4315

Framework versions

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