metadata
language:
- de
license: apache-2.0
library_name: peft
tags:
- generated_from_trainer
base_model: openai/whisper-small
datasets:
- rmacek/ORF-whisper-small
metrics:
- wer
model-index:
- name: Whisper ORF Bundeslaender
results:
- task:
type: automatic-speech-recognition
name: Automatic Speech Recognition
dataset:
name: ZIB2 Common Voice
type: rmacek/ORF-whisper-small
args: 'config: de, split: test'
metrics:
- type: wer
value: 29.46806728721495
name: Wer
Whisper ORF Bundeslaender
This model is a fine-tuned version of openai/whisper-small on the ZIB2 Common Voice dataset. It achieves the following results on the evaluation set:
- Loss: 0.7149
- Wer: 29.4681
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: 0.0001
- train_batch_size: 16
- eval_batch_size: 8
- seed: 42
- optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
- lr_scheduler_type: linear
- lr_scheduler_warmup_steps: 500
- training_steps: 8000
- mixed_precision_training: Native AMP
Training results
| Training Loss | Epoch | Step | Validation Loss | Wer |
|---|---|---|---|---|
| 0.5717 | 1.7153 | 1000 | 0.6098 | 35.2851 |
| 0.41 | 3.4305 | 2000 | 0.6053 | 32.7056 |
| 0.386 | 5.1458 | 3000 | 0.6148 | 26.9699 |
| 0.3123 | 6.8611 | 4000 | 0.6345 | 26.4842 |
| 0.2183 | 8.5763 | 5000 | 0.6622 | 28.1329 |
| 0.2231 | 10.2916 | 6000 | 0.6901 | 28.5253 |
| 0.2259 | 12.0069 | 7000 | 0.7028 | 27.4413 |
| 0.1708 | 13.7221 | 8000 | 0.7149 | 29.4681 |
Framework versions
- PEFT 0.10.1.dev0
- Transformers 4.40.0
- Pytorch 2.2.1+cu121
- Datasets 2.19.0
- Tokenizers 0.19.1