hip94/52Hz-tiny-fr-v2

52Hz Tiny Fr - IMT Atlantique X 52 Hertz

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

• Loss: 0.6781
• Wer: 78.6424

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
• 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
4.3806	1.0	21	1.7951	935.5960
2.2732	2.0	42	1.1379	211.4238
1.5989	3.0	63	0.9543	159.4371
1.1012	4.0	84	0.8132	111.0927
0.9105	5.0	105	0.7807	57.7815
0.6273	6.0	126	0.7140	88.0795
0.6452	7.0	147	0.7138	71.5232
0.5428	8.0	168	0.6825	80.9603
0.4051	9.0	189	0.6651	79.9669
0.3947	10.0	210	0.6726	79.4702
0.3583	11.0	231	0.6810	77.1523
0.3139	12.0	252	0.6745	77.8146
0.3343	13.0	273	0.6796	67.0530
0.3315	14.0	294	0.6787	78.8079
0.2934	15.0	315	0.6781	78.6424

Framework versions

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