transcription_full

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

• Loss: 0.0528
• Wer: 0.4930
• Cer: 0.2717

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 4
• 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: cosine
• lr_scheduler_warmup_steps: 100
• num_epochs: 4
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
0.0917	0.2531	4200	0.3473	0.0922	0.6812
0.078	0.5063	8400	0.3166	0.0766	0.5952
0.0689	0.7594	12600	0.2966	0.0691	0.5526
0.062	1.0126	16800	0.2926	0.0643	0.5457
0.0583	1.2657	21000	0.2725	0.0611	0.5045
0.0597	1.5189	25200	0.0592	0.5154	0.2810
0.0504	1.7720	29400	0.0577	0.5082	0.2784
0.0481	2.0252	33600	0.0565	0.4934	0.2707
0.049	2.2783	37800	0.0549	0.4839	0.2662
0.0483	2.5315	42000	0.0537	0.4911	0.2712
0.0495	2.7846	46200	0.0528	0.4930	0.2717

Framework versions

• PEFT 0.18.0
• Transformers 4.57.3
• Pytorch 2.9.0+cu130
• Datasets 2.21.0
• Tokenizers 0.22.1