MathildeB3/52Hz-small-fr-v2b

52Hz Small Fr - IMT Atlantique X 52 Hertz

This model is a fine-tuned version of openai/whisper-small on the Transcriptions IMTx52Hz v2 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4081
• Wer: 17.4837

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: 2e-05
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• 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: cosine
• lr_scheduler_warmup_ratio: 0.1
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.4611	1.0	22	0.8951	40.3595
0.6363	2.0	44	0.5120	25.1634
0.2988	3.0	66	0.4660	20.4248
0.2174	4.0	88	0.4379	36.7647
0.1602	5.0	110	0.4227	44.9346
0.1294	6.0	132	0.4130	18.6275
0.0972	7.0	154	0.4207	18.1373
0.0591	8.0	176	0.4045	19.4444
0.0377	9.0	198	0.4233	15.6863
0.0337	10.0	220	0.4078	17.1569
0.0292	11.0	242	0.4063	17.6471
0.0225	12.0	264	0.4063	16.9935
0.0313	13.0	286	0.4073	17.4837
0.0185	14.0	308	0.4080	17.4837
0.0198	15.0	330	0.4081	17.4837

Framework versions

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