metadata
library_name: peft
language:
- it
license: cc-by-nc-4.0
base_model: nyrahealth/CrisperWhisper
tags:
- generated_from_trainer
datasets:
- b-brave-clean
metrics:
- wer
model-index:
- name: Whisper-Crisper
results:
- task:
type: automatic-speech-recognition
name: Automatic Speech Recognition
dataset:
name: b-brave-clean
type: b-brave-clean
config: default
split: test
args: default
metrics:
- type: wer
value: 155.63816604708796
name: Wer
Whisper-Crisper
This model is a fine-tuned version of nyrahealth/CrisperWhisper on the b-brave-clean dataset. It achieves the following results on the evaluation set:
- Loss: 0.2971
- Wer: 155.6382
- Cer: 74.4218
- 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: 4
- eval_batch_size: 4
- seed: 42
- gradient_accumulation_steps: 2
- total_train_batch_size: 8
- 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: 0.5
- num_epochs: 8
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer | Cer | Lr |
|---|---|---|---|---|---|---|
| 1.0227 | 1.0 | 335 | 0.8979 | 601.9827 | 363.8095 | 0.0001 |
| 0.5552 | 2.0 | 670 | 0.5294 | 1520.4461 | 678.7302 | 0.0002 |
| 0.4207 | 3.0 | 1005 | 0.4611 | 403.5936 | 219.1610 | 0.0002 |
| 0.2025 | 4.0 | 1340 | 0.3812 | 346.8401 | 162.0408 | 0.0003 |
| 0.1216 | 5.0 | 1675 | 0.3001 | 400.3717 | 203.6054 | 0.0002 |
| 0.0478 | 6.0 | 2010 | 0.2932 | 198.6369 | 94.8299 | 0.0001 |
| 0.0313 | 7.0 | 2345 | 0.3033 | 241.0161 | 115.3288 | 0.0001 |
| 0.012 | 7.9776 | 2672 | 0.2971 | 155.6382 | 74.4218 | 0.0000 |
Framework versions
- PEFT 0.14.0
- Transformers 4.48.3
- Pytorch 2.2.0
- Datasets 3.2.0
- Tokenizers 0.21.0