zuazo/whisper-large-v3-eu-train

Whisper Large-V3 Basque

This model is a fine-tuned version of openai/whisper-large-v3 on the mozilla-foundation/common_voice_13_0 eu dataset. It achieves the following results on the evaluation set:

• Loss: 0.4180
• Wer: 13.2886

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: 2
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 20000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.1288	5.85	1000	0.2746	18.6000
0.0262	11.7	2000	0.2894	16.0934
0.0095	17.54	3000	0.3281	15.7348
0.0056	23.39	4000	0.3362	14.7394
0.0045	29.24	5000	0.3465	14.9912
0.0032	35.09	6000	0.3599	14.7172
0.002	40.94	7000	0.3624	14.4150
0.0028	46.78	8000	0.3647	14.4553
0.0019	52.63	9000	0.3726	14.4210
0.0011	58.48	10000	0.3784	14.1268
0.0011	64.33	11000	0.3753	14.2517
0.0009	70.18	12000	0.3845	13.9193
0.0008	76.02	13000	0.3910	14.0402
0.0008	81.87	14000	0.3988	13.8488
0.0004	87.72	15000	0.4002	13.5788
0.0002	93.57	16000	0.4021	13.5526
0.0002	99.42	17000	0.4121	13.5747
0.0002	105.26	18000	0.4178	13.5989
0.0005	111.11	19000	0.4135	13.3551
0.0001	116.96	20000	0.4180	13.2886

Framework versions

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