metadata

library_name: transformers
language:
  - ar
license: apache-2.0
base_model: openai/whisper-large
tags:
  - automatic-speech-recognition
  - arabic
  - generated_from_trainer
datasets:
  - audiofolder
metrics:
  - wer
model-index:
  - name: Whisper Large Informal Arabic
    results:
      - task:
          type: automatic-speech-recognition
          name: Automatic Speech Recognition
        dataset:
          name: Informal Arabic
          type: audiofolder
          config: default
          split: None
          args: default
        metrics:
          - type: wer
            value: 24.96401151631478
            name: Wer

Whisper Large Informal Arabic

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

Loss: 0.5264
Wer: 24.9640
Cer: 8.1265

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-05
train_batch_size: 2
eval_batch_size: 2
seed: 42
gradient_accumulation_steps: 8
total_train_batch_size: 16
optimizer: Use OptimizerNames.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: 250
training_steps: 2500
mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer	Cer
0.0054	13.1611	500	0.4210	27.3153	9.2113
0.0002	26.3221	1000	0.4803	24.9640	7.9997
0.0001	39.4832	1500	0.5063	24.6881	7.9997
0.0001	52.6443	2000	0.5200	24.7001	8.0326
0.0001	65.8054	2500	0.5264	24.9640	8.1265

Framework versions

Transformers 4.50.3
Pytorch 2.6.0+cu124
Datasets 3.5.0
Tokenizers 0.21.1