whisper-base.hr

This model is a fine-tuned version of openai/whisper-base on the GoranS/stt-croatian_99k_265_2 dataset. It achieves the following results on the evaluation set:

• Loss: 0.3545
• Wer: 0.2541

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: 6.25e-06
• train_batch_size: 64
• eval_batch_size: 32
• seed: 42
• 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: 800
• num_epochs: 5
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.6206	0.5450	1000	0.5805	0.4175
0.4639	1.0899	2000	0.4649	0.3324
0.3994	1.6349	3000	0.4169	0.2988
0.3571	2.1798	4000	0.3907	0.2884
0.3248	2.7248	5000	0.3748	0.2790
0.2987	3.2698	6000	0.3660	0.2726
0.3133	3.8147	7000	0.3589	0.2544
0.2829	4.3597	8000	0.3565	0.2629
0.2688	4.9046	9000	0.3545	0.2541

Framework versions

• Transformers 4.47.0
• Pytorch 2.5.1+cu124
• Datasets 3.1.0
• Tokenizers 0.21.0