arun100/whisper-small-fr-derived-1

Whisper Base French

This model is a fine-tuned version of qanastek/whisper-small-french-uncased on the mozilla-foundation/common_voice_16_0 fr dataset. It achieves the following results on the evaluation set:

• Loss: 0.8014
• Wer: 15.1845

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: 5e-07
• train_batch_size: 16
• eval_batch_size: 32
• seed: 42
• gradient_accumulation_steps: 4
• total_train_batch_size: 64
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.9295	0.2	100	0.8014	15.1845
0.2976	0.4	200	0.4207	16.0289
0.2699	0.59	300	0.3999	15.8267
0.2773	0.79	400	0.3910	15.7267
0.2631	0.99	500	0.3863	15.5972
0.2487	1.19	600	0.3834	15.5907
0.2477	1.39	700	0.3814	15.6156
0.2428	1.59	800	0.3801	15.4902
0.2492	1.78	900	0.3794	15.4672
0.2471	1.98	1000	0.3791	15.4707

Framework versions

• Transformers 4.38.0.dev0
• Pytorch 2.1.2+cu121
• Datasets 2.16.2.dev0
• Tokenizers 0.15.0