personalizedrefrigerator/whisper-tiny-fr

Whisper Tiny (Finetuned on French)

This model is a fine-tuned version of openai/whisper-tiny on the Common Voice 11.0 dataset. It achieves the following results on the evaluation set:

• Loss: 0.7793
• Model Preparation Time: 0.0027
• Wer: 0.4218
• Cer: 0.1940

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: 16
• eval_batch_size: 8
• seed: 42
• 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
• training_steps: 12000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Wer	Cer
0.516	0.0833	1000	0.9995	0.0027	0.4487	0.2213
0.5283	0.1667	2000	0.9679	0.0027	0.4511	0.2207
0.5421	0.25	3000	0.9532	0.0027	0.4462	0.2172
0.5735	0.3333	4000	0.9474	0.0027	0.4365	0.2110
0.5774	0.4167	5000	0.9119	0.0027	0.4794	0.2390
0.591	0.5	6000	0.8834	0.0027	0.4171	0.2024
0.5218	0.5833	7000	0.8777	0.0027	0.4293	0.2096
0.4328	0.6667	8000	0.8750	0.0027	0.4139	0.2017
0.5392	0.75	9000	0.8736	0.0027	0.5618	0.3050
0.4311	0.8333	10000	0.8587	0.0027	0.5618	0.3030
0.4728	0.9167	11000	0.8514	0.0027	0.4293	0.2034
0.4521	1.0	12000	0.8516	0.0027	0.4220	0.2054

Framework versions

• Transformers 4.46.3
• Pytorch 2.5.1+cu121
• Datasets 3.1.0
• Tokenizers 0.20.3