zuazo/whisper-tiny-pt

Whisper Tiny Portuguese

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

• Loss: 0.9091
• Wer: 28.2309

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: 3.75e-05
• train_batch_size: 256
• eval_batch_size: 128
• 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
0.0172	14.08	1000	0.6518	29.2084
0.0054	28.17	2000	0.7505	29.3431
0.0018	42.25	3000	0.7833	28.7944
0.0008	56.34	4000	0.8186	28.9702
0.0012	70.42	5000	0.8409	29.6241
0.0004	84.51	6000	0.8530	29.0014
0.0002	98.59	7000	0.8743	28.3163
0.0001	112.68	8000	0.8918	28.2506
0.0001	126.76	9000	0.9091	28.2309
0.0001	140.85	10000	0.9273	28.4001
0.0	154.93	11000	0.9460	28.6005
0.0	169.01	12000	0.9660	28.5496
0.0	183.1	13000	0.9863	28.5151
0.0	197.18	14000	1.0075	28.5660
0.0	211.27	15000	1.0271	28.5907
0.0	225.35	16000	1.0469	28.5003
0.0	239.44	17000	1.0647	28.5200
0.0	253.52	18000	1.0800	28.5036
0.0	267.61	19000	1.0910	28.5545
0.0	281.69	20000	1.0960	28.6071

Framework versions

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