v3-turbo-low-lora-8805-qkvo

This model is a fine-tuned version of openai/whisper-large-v3-turbo on the Cagri-kayitlar-relabeled-06s-200ms-padded dataset. It achieves the following results on the evaluation set:

• Loss: 0.3629
• Wer: 28.7627

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: 4
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• 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: cosine
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.1764	1.0482	500	0.4268	32.1961
0.3681	2.0964	1000	0.3905	31.4835
0.3222	3.1447	1500	0.3774	30.5981
0.3158	4.1929	2000	0.3698	29.2377
0.2983	5.2411	2500	0.3670	28.9786
0.283	6.2893	3000	0.3649	28.8059
0.2714	7.3375	3500	0.3628	28.4172
0.263	8.3857	4000	0.3629	28.5468
0.2656	9.4340	4500	0.3629	28.7627
0.2607	10.4822	5000	0.3629	28.7627

Framework versions

• PEFT 0.14.0
• Transformers 4.48.3
• Pytorch 2.4.1+cu121
• Datasets 3.1.0
• Tokenizers 0.21.0