metadata
library_name: peft
language:
- it
license: apache-2.0
base_model: miosipof/whisper-medium-it-luigisaetta
tags:
- generated_from_trainer
datasets:
- b-brave-balanced-augmented
metrics:
- wer
model-index:
- name: Whisper Medium IT
results:
- task:
type: automatic-speech-recognition
name: Automatic Speech Recognition
dataset:
name: b-brave-balanced-augmented
type: b-brave-balanced-augmented
metrics:
- type: wer
value: 33.11081441922563
name: Wer
Whisper Medium IT
This model is a fine-tuned version of openai/whisper-medium on the b-brave-balanced-augmented dataset. It achieves the following results on the evaluation set:
- Loss: 0.2996
- Wer: 33.1108
- Cer: 21.3778
- Lr: 0.0000
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: 0.0003
- train_batch_size: 50
- eval_batch_size: 50
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 100
- 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_ratio: 0.3
- num_epochs: 4
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer | Lr |
|---|---|---|---|---|---|---|
| 12.0253 | 1.0 | 83 | 1.0195 | 72.8972 | 40.7434 | 0.0002 |
| 1.5523 | 2.0 | 166 | 0.5319 | 50.0668 | 31.6288 | 0.0002 |
| 1.0161 | 3.0 | 249 | 0.3637 | 40.1869 | 25.7102 | 0.0001 |
| 0.5605 | 4.0 | 332 | 0.2996 | 33.1108 | 21.3778 | 0.0000 |
Framework versions
- PEFT 0.15.1
- Transformers 4.47.1
- Pytorch 2.2.0
- Datasets 3.5.0
- Tokenizers 0.21.1