whisper-medium-bigcgen-combined-10hrs-model

This model is a fine-tuned version of openai/whisper-medium on the bigcgen dataset. It achieves the following results on the evaluation set:

• Loss: 0.6296
• Wer: 0.5246

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: 2
• eval_batch_size: 2
• seed: 42
• gradient_accumulation_steps: 4
• 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: 500
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
3.7039	0.3048	200	0.9005	0.6247
3.2235	0.6095	400	0.7650	0.5821
3.1161	0.9143	600	0.7064	0.6293
2.3214	1.2179	800	0.6911	0.5950
2.0726	1.5227	1000	0.6554	0.5229
1.9852	1.8274	1200	0.6296	0.5246
1.2967	2.1310	1400	0.6494	0.4759
1.3477	2.4358	1600	0.6510	0.5085
1.3943	2.7406	1800	0.6315	0.4910

Framework versions

• Transformers 4.47.1
• Pytorch 2.5.1+cu124
• Datasets 3.2.0
• Tokenizers 0.21.0