Medical Whisper - Portuguese

This model is a fine-tuned version of openai/whisper-large-v3-turbo on an unknown dataset. It achieves the following results on the evaluation set:

  • Loss: 0.3011
  • Wer: 30.6945

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: 9e-06
  • train_batch_size: 4
  • eval_batch_size: 16
  • seed: 42
  • distributed_type: multi-GPU
  • num_devices: 4
  • gradient_accumulation_steps: 8
  • total_train_batch_size: 128
  • total_eval_batch_size: 64
  • optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
  • lr_scheduler_type: linear
  • lr_scheduler_warmup_steps: 500
  • training_steps: 6000
  • mixed_precision_training: Native AMP

Training results

Training Loss Epoch Step Validation Loss Wer
0.3654 0.3680 500 0.3516 35.6868
0.3355 0.7360 1000 0.3341 35.0002
0.2826 1.1041 1500 0.3248 34.6635
0.2763 1.4721 2000 0.3171 33.6200
0.2715 1.8401 2500 0.3101 33.1267
0.2203 2.2081 3000 0.3071 31.3256
0.2202 2.5761 3500 0.3019 30.5031
0.2169 2.9442 4000 0.2975 30.7246
0.1765 3.3122 4500 0.3002 31.2968
0.1768 3.6802 5000 0.2985 30.5046
0.1594 4.0482 5500 0.3003 30.5781
0.1603 4.4162 6000 0.3011 30.6945

Framework versions

  • Transformers 4.46.0.dev0
  • Pytorch 2.1.0+cu118
  • Datasets 3.0.2.dev0
  • Tokenizers 0.20.0
Downloads last month
26
Safetensors
Model size
809M params
Tensor type
F32
·
Inference Examples
This model does not have enough activity to be deployed to Inference API (serverless) yet. Increase its social visibility and check back later, or deploy to Inference Endpoints (dedicated) instead.

Model tree for todeschini/medical-whisper-pt

Finetuned
(125)
this model