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: 0.4799
• Wer: 6.3993

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-06
• 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
0.077	3.53	1000	0.1616	5.4957
0.0155	7.05	2000	0.2549	6.1956
0.0045	10.58	3000	0.3122	5.9261
0.0017	14.11	4000	0.3317	6.0099
0.0018	17.64	5000	0.3604	6.0099
0.0009	21.16	6000	0.3779	6.1791
0.0012	24.69	7000	0.3470	6.0066
0.0013	28.22	8000	0.3838	6.1479
0.0007	31.75	9000	0.3839	6.0395
0.0003	35.27	10000	0.4090	6.2054
0.0003	38.8	11000	0.4053	6.2859
0.0002	42.33	12000	0.4235	6.3467
0.0002	45.86	13000	0.4326	6.3500
0.0001	49.38	14000	0.4415	6.3714
0.0001	52.91	15000	0.4506	6.3878
0.0001	56.44	16000	0.4586	6.4092
0.0001	59.96	17000	0.4663	6.3944
0.0001	63.49	18000	0.4730	6.3911
0.0001	67.02	19000	0.4778	6.3944
0.0001	70.55	20000	0.4799	6.3993

Framework versions

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