zuazo/whisper-large-eu-from-es

Whisper Large Basque

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

• Loss: 0.4549
• Wer: 12.9815

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

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0162	4.01	1000	0.3159	16.8577
0.0042	9.01	2000	0.3181	15.1606
0.0038	14.01	3000	0.3367	14.7211
0.0035	19.0	4000	0.3419	14.5915
0.0012	24.0	5000	0.3489	14.3586
0.0029	29.0	6000	0.3650	14.6746
0.0011	33.01	7000	0.3643	13.8138
0.0006	38.01	8000	0.3628	14.0042
0.0009	43.01	9000	0.3661	14.0042
0.0003	48.01	10000	0.3794	13.7166
0.0003	53.0	11000	0.3793	13.6923
0.0	58.0	12000	0.3991	13.4027
0.0	63.0	13000	0.4119	13.3562
0.0	67.01	14000	0.4209	13.2184
0.0	72.01	15000	0.4288	13.2225
0.0	77.01	16000	0.4361	13.1516
0.0	82.01	17000	0.4428	13.1334
0.0	87.0	18000	0.4487	13.1334
0.0	92.0	19000	0.4531	12.9896
0.0	97.0	20000	0.4549	12.9815

Framework versions

• Transformers 4.33.0.dev0
• Pytorch 2.0.1+cu117
• Datasets 2.14.4
• Tokenizers 0.13.3