whisper-large-v2-phase2-test

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: 0.5457
• Cer: 16.3233
• Wer: 27.7114

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 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
1.1099	1.0	2111	0.4675	25.1766	40.3020
0.6578	2.0	4222	0.4679	21.6808	35.4688
0.5034	3.0	6333	0.4749	21.5618	35.0969
0.3984	4.0	8444	0.4755	23.2471	38.2574
0.3204	5.0	10555	0.4918	19.0426	31.4930
0.2566	6.0	12666	0.4970	18.2685	30.2707
0.2062	7.0	14777	0.5151	18.2196	30.6263
0.1668	8.0	16888	0.5326	16.6726	28.2915
0.1376	9.0	18999	0.5457	16.3233	27.7114
0.1107	10.0	21110	0.5543	16.8117	28.3746
0.0932	11.0	23221	0.5640	16.6303	28.1617
0.0789	12.0	25332	0.5871	16.5051	28.0401
0.0671	13.0	27443	0.5943	16.3639	27.7605
0.0596	14.0	29554	0.6002	16.5290	28.0061

Framework versions

• Transformers 4.57.1
• Pytorch 2.9.0+cu128
• Datasets 3.6.0
• Tokenizers 0.22.1