alpcansoydas/whisper-large-v2-tr-ft-26-03-26-encoder-only-25ksamples-simulated-data

Whisper for Turkish Call Centers

This model is a fine-tuned version of openai/whisper-large-v2 on the Custom turkish call center simulated data dataset. It achieves the following results on the evaluation set:

• Loss: 0.2845

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-06
• train_batch_size: 16
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 32
• 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: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 4500
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss
0.6040	0.3554	250	0.3119
0.5851	0.7107	500	0.2891
0.5170	1.0654	750	0.2810
0.5519	1.4208	1000	0.2746
0.6250	1.7761	1250	0.2717
0.4590	2.1308	1500	0.2713
0.4245	2.4861	1750	0.2685
0.4649	2.8415	2000	0.2674
0.3951	3.1962	2250	0.2712
0.4086	3.5515	2500	0.2694
0.4263	3.9069	2750	0.2694
0.3639	4.2615	3000	0.2739
0.4839	4.6169	3250	0.2739
0.3619	4.9723	3500	0.2740
0.3577	5.3269	3750	0.2800
0.3624	5.6823	4000	0.2804
0.3432	6.0370	4250	0.2823
0.3676	6.3923	4500	0.2845

Framework versions

• Transformers 5.3.0
• Pytorch 2.10.0+cu128
• Datasets 4.8.4
• Tokenizers 0.22.2