whisper-large-v3-turbo-ESLO_REP

This model is a fine-tuned version of openai/whisper-large-v3-turbo on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 1.3035
• Wer: 65.8438

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: 1e-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 with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• num_epochs: 15
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.0684	1.0	299	0.9910	67.5694
0.8539	2.0	598	0.9916	74.2320
0.6497	3.0	897	0.9486	61.4563
0.4445	4.0	1196	0.9633	70.8195
0.3109	5.0	1495	1.0225	86.3886
0.2237	6.0	1794	1.0533	68.8076
0.1556	7.0	2093	1.0990	66.6099
0.1219	8.0	2392	1.1393	65.8671
0.0916	9.0	2691	1.1718	67.8558
0.069	10.0	2990	1.1896	66.9659
0.0592	11.0	3289	1.2018	69.3957
0.0517	12.0	3588	1.2355	66.6022
0.0416	13.0	3887	1.2457	66.2230
0.0335	14.0	4186	1.2507	65.9290
0.0261	15.0	4485	1.3035	65.8438

Framework versions

• Transformers 4.46.2
• Pytorch 2.4.1+cu121
• Datasets 3.1.0
• Tokenizers 0.20.0