whisper-tiny-no-specific-topic-v3

This model is a fine-tuned version of openai/whisper-tiny on the None dataset. It achieves the following results on the evaluation set:

• Loss: 0.8726
• Wer: 40.7091

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: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 16
• 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: 500
• training_steps: 8000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4891	0.125	1000	0.7647	45.5818
0.4681	0.25	2000	0.7404	45.5727
0.2846	0.375	3000	0.7901	41.1909
0.3004	0.5	4000	0.8354	45.9455
0.2533	0.625	5000	0.8392	41.8182
0.2314	0.75	6000	0.8626	41.5
0.2622	0.875	7000	0.8690	43.5545
0.2165	1.0	8000	0.8726	40.7091

Framework versions

• Transformers 4.49.0
• Pytorch 2.5.1+cu121
• Datasets 3.3.2
• Tokenizers 0.21.0