zuazo/whisper-large-pt

Whisper Large Portuguese

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

• Loss: 0.4317
• Wer: 7.4196

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
0.0299	3.53	1000	0.1977	6.2416
0.0071	7.05	2000	0.2617	6.7805
0.0055	10.58	3000	0.2787	7.3670
0.0024	14.11	4000	0.3019	7.2717
0.0034	17.64	5000	0.3056	7.4902
0.0016	21.16	6000	0.3180	7.5691
0.0021	24.69	7000	0.3234	7.7432
0.0028	28.22	8000	0.3293	7.9848
0.0003	31.75	9000	0.3540	7.8944
0.0003	35.27	10000	0.3439	7.8862
0.0005	38.8	11000	0.3673	8.1704
0.0008	42.33	12000	0.3510	7.9371
0.0001	45.86	13000	0.3695	7.8895
0.0	49.38	14000	0.3879	7.6151
0.0	52.91	15000	0.3993	7.5461
0.0	56.44	16000	0.4087	7.5116
0.0	59.96	17000	0.4170	7.4672
0.0	63.49	18000	0.4241	7.4344
0.0	67.02	19000	0.4294	7.4179
0.0	70.55	20000	0.4317	7.4196

Framework versions

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