hip94/52Hz-small-fr-v2

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: 0.2991
• Wer: 10.0993

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
1.6508	1.0	21	0.7795	28.8079
0.6567	2.0	42	0.3693	18.0464
0.3608	3.0	63	0.2880	13.5762
0.2693	4.0	84	0.2782	13.2450
0.1341	5.0	105	0.2645	12.4172
0.1078	6.0	126	0.2908	13.4106
0.0887	7.0	147	0.2834	12.9139
0.0626	8.0	168	0.2862	10.5960
0.0461	9.0	189	0.2833	10.9272
0.0283	10.0	210	0.2834	10.4305
0.0333	11.0	231	0.2935	9.4371
0.0219	12.0	252	0.2925	10.4305
0.0265	13.0	273	0.3023	10.2649
0.0274	14.0	294	0.2992	10.0993
0.0216	15.0	315	0.2991	10.0993

Framework versions

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