hip94/whisper-tiny-fr

52Hz Tiny Fr - IMT Atlantique X 52 Hertz

This model is a fine-tuned version of openai/whisper-tiny on the Premier dataset organisé de 52 Hertz dataset. It achieves the following results on the evaluation set:

• Loss: 0.6664
• Wer: 58.6381

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-05
• train_batch_size: 16
• eval_batch_size: 16
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• optimizer: Use OptimizerNames.ADAMW_TORCH_FUSED with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
5.0278	1.0	12	1.9196	763.3039
3.4923	2.0	24	1.4318	492.8121
1.9181	3.0	36	1.0781	178.0580
1.5961	4.0	48	0.9307	100.1261
1.2186	5.0	60	0.8604	93.4426
0.9392	6.0	72	0.8176	68.8525
0.8361	7.0	84	0.7492	38.7137
0.783	8.0	96	0.7197	69.2308
0.68	9.0	108	0.6915	40.7314
0.6685	10.0	120	0.6762	56.3682
0.5406	11.0	132	0.6753	63.0517
0.5741	12.0	144	0.6721	59.6469
0.5247	13.0	156	0.6677	62.9256
0.5278	14.0	168	0.6663	62.0429
0.4853	15.0	180	0.6664	58.6381

Framework versions

• Transformers 4.57.3
• Pytorch 2.9.1+cu130
• Datasets 4.4.2
• Tokenizers 0.22.2