whisper-large-v3-natbed-non-native-model

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: 0.8693
• Wer: 52.1349

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: 1.75e-05
• train_batch_size: 4
• eval_batch_size: 4
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 8
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 30000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
1.5641	0.5631	250	0.8395	70.9929
0.821	1.1261	500	0.7848	65.2830
0.6791	1.6892	750	0.7238	59.8611
0.5596	2.2523	1000	0.7156	55.1339
0.46	2.8153	1250	0.7180	54.4222
0.3263	3.3784	1500	0.7762	56.3707
0.3088	3.9414	1750	0.7282	51.8807
0.1838	4.5045	2000	0.7987	52.5246
0.1694	5.0676	2250	0.8901	53.5920
0.1054	5.6306	2500	0.8693	52.1349

Framework versions

• Transformers 4.45.0.dev0
• Pytorch 2.4.1+cu121
• Datasets 3.0.0
• Tokenizers 0.19.1