whisper_large_v3

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: 0.5667
• Cer: 23.6266
• Wer: 37.1181

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• 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: 1000
• num_epochs: 10
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Cer	Validation Loss	Wer
2.4223	1.0	1056	43.3279	0.7348	66.1582
0.8323	2.0	2112	43.1177	0.6394	63.7241
0.5984	3.0	3168	35.8693	0.5945	54.4320
0.4851	4.0	4224	37.6694	0.5898	55.6333
0.4166	5.0	5280	32.2482	0.5693	48.2993
0.4257	6.0	6336	0.5744	33.5367	50.8352
0.3691	7.0	7392	0.5659	28.8876	43.8848
0.3243	8.0	8448	0.5641	23.7833	37.6421
0.2896	9.0	9504	0.5677	23.8392	37.5977
0.2669	10.0	10560	0.5667	23.6266	37.1181

Framework versions

• Transformers 4.55.4
• Pytorch 2.8.0+cu126
• Datasets 4.1.1
• Tokenizers 0.21.4