MathildeB3/52Hz-small-fr

52Hz Small Fr - IMT Atlantique X 52 Hertz

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

• Loss: 0.5492
• Wer: 45.9384

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• 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: constant_with_warmup
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.4736	1.0	23	1.4884	104.3417
1.2019	2.0	46	1.1218	60.5042
0.8634	3.0	69	0.8382	55.7423
0.593	4.0	92	0.6966	52.2409
0.4493	5.0	115	0.6234	50.8403
0.3896	6.0	138	0.5908	50.2801
0.3281	7.0	161	0.5737	47.6190
0.2867	8.0	184	0.5482	50.5602
0.2528	9.0	207	0.5397	47.1989
0.2379	10.0	230	0.5455	47.4790
0.1741	11.0	253	0.5469	47.1989
0.1718	12.0	276	0.5458	47.1989
0.1213	13.0	299	0.5413	45.3782
0.1177	14.0	322	0.5450	46.2185
0.0879	15.0	345	0.5492	45.9384

Framework versions

• PEFT 0.18.1
• Transformers 4.57.3
• Pytorch 2.9.1+cu130
• Datasets 4.4.2
• Tokenizers 0.22.2