dhaval0108/whisper-large-v3-turbo-finetuned-frozen-encoder

ap-LracgWanIBIvdjeHeDoO07

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.5750
• Model Preparation Time: 0.025
• Wer: 0.1411

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: 0.0003
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 8
• total_train_batch_size: 64
• 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: 400
• num_epochs: 16
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Model Preparation Time	Wer
0.3835	1.0	250	0.4302	0.025	0.2568
0.3373	2.0	500	0.4448	0.025	0.1976
0.2796	3.0	750	0.3961	0.025	0.1652
0.2106	4.0	1000	0.3875	0.025	0.1811
0.1608	5.0	1250	0.3860	0.025	0.1693
0.1203	6.0	1500	0.3717	0.025	0.1520
0.0948	7.0	1750	0.3979	0.025	0.1592
0.0723	8.0	2000	0.4183	0.025	0.1511
0.0569	9.0	2250	0.4161	0.025	0.1597
0.0423	10.0	2500	0.4580	0.025	0.1540
0.031	11.0	2750	0.4652	0.025	0.1389
0.0208	12.0	3000	0.4817	0.025	0.1391
0.0162	13.0	3250	0.5002	0.025	0.1369
0.0087	14.0	3500	0.5394	0.025	0.1378
0.0033	15.0	3750	0.5667	0.025	0.1400
0.0013	15.9365	3984	0.5750	0.025	0.1411

Framework versions

• Transformers 4.48.3
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.2