whisper-large-v2-Split-Sentences

This model is a fine-tuned version of openai/whisper-large-v2 on an unknown dataset. It achieves the following results on the evaluation set:

• Loss: 0.5782
• Cer: 15.1758
• Wer: 26.5014

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: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Use 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: 1000
• num_epochs: 25
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Cer	Wer
1.2919	1.0	1353	0.5490	23.5550	38.4031
0.7394	2.0	2706	0.5212	19.1820	32.3358
0.5884	3.0	4059	0.5093	19.2099	31.6647
0.4839	4.0	5412	0.5118	21.1652	34.1094
0.4026	5.0	6765	0.5198	18.4545	29.5624
0.3374	6.0	8118	0.5287	16.1245	27.6039
0.2818	7.0	9471	0.5281	16.6447	28.2682
0.2356	8.0	10824	0.5510	15.9731	27.0561
0.1961	9.0	12177	0.5678	15.9691	27.4190
0.164	10.0	13530	0.5782	15.1758	26.5014
0.1378	11.0	14883	0.5843	15.3113	26.2001
0.1149	12.0	16236	0.6060	16.0169	27.2341
0.0959	13.0	17589	0.6150	15.9671	27.4053
0.0808	14.0	18942	0.6438	15.4768	26.3165
0.0675	15.0	20295	0.6502	15.5605	26.0563

Framework versions

• Transformers 4.53.3
• Pytorch 2.7.1+cu118
• Datasets 3.6.0
• Tokenizers 0.21.2