zuazo/whisper-small-pt

Whisper Small Portuguese

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

• Loss: 0.2530
• Wer: 10.1912

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: 64
• eval_batch_size: 32
• 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: 20000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.0522	3.52	1000	0.2530	10.1912
0.008	7.04	2000	0.3280	10.3161
0.0038	10.56	3000	0.3617	10.5165
0.0024	14.08	4000	0.3780	10.7498
0.001	17.61	5000	0.4050	10.5872
0.0019	21.13	6000	0.3971	10.9700
0.001	24.65	7000	0.4287	10.9634
0.0009	28.17	8000	0.4292	11.1934
0.0019	31.69	9000	0.4331	11.0554
0.0005	35.21	10000	0.4535	11.0604
0.0012	38.73	11000	0.4365	11.2822
0.0003	42.25	12000	0.4511	11.0866
0.0001	45.77	13000	0.4710	10.9256
0.0001	49.3	14000	0.4822	10.9191
0.0001	52.82	15000	0.4916	10.8796
0.0	56.34	16000	0.5005	10.8879
0.0	59.86	17000	0.5088	10.9224
0.0	63.38	18000	0.5159	10.9256
0.0	66.9	19000	0.5215	10.9092
0.0	70.42	20000	0.5240	10.9191

Framework versions

• Transformers 4.37.2
• Pytorch 2.0.1+cu117
• Datasets 2.14.4
• Tokenizers 0.15.1