zuazo/whisper-large-v3-pt-old

Whisper Large-V3 Portuguese

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

• Loss: 0.3553
• Wer: 5.1802

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
0.0316	3.53	1000	0.1632	4.6397
0.0066	7.05	2000	0.2164	4.8944
0.0032	10.58	3000	0.2438	5.0619
0.0011	14.11	4000	0.2523	5.0751
0.0037	17.64	5000	0.2481	5.2460
0.0011	21.16	6000	0.2683	5.3692
0.0033	24.69	7000	0.2756	5.5844
0.0009	28.22	8000	0.2769	5.4628
0.0013	31.75	9000	0.2664	5.4349
0.0007	35.27	10000	0.3020	5.4776
0.0005	38.8	11000	0.2886	5.4595
0.0003	42.33	12000	0.3016	5.3265
0.0003	45.86	13000	0.3040	5.5121
0.0001	49.38	14000	0.3147	5.4480
0.0001	52.91	15000	0.3071	5.4300
0.0	56.44	16000	0.3307	5.3051
0.0	59.96	17000	0.3412	5.2476
0.0	63.49	18000	0.3483	5.2016
0.0	67.02	19000	0.3532	5.1884
0.0	70.55	20000	0.3553	5.1802

Framework versions

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