whisper-medium-bigcgen-combined-15hrs-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: 0.5868
• Wer: 0.4709

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
3.7191	0.2035	200	0.9003	0.6238
3.3861	0.4070	400	0.7572	0.5558
2.9296	0.6105	600	0.7156	0.5460
2.8891	0.8140	800	0.6542	0.5140
2.0307	1.0183	1000	0.6246	0.5156
2.277	1.2218	1200	0.6171	0.5276
1.8914	1.4253	1400	0.6088	0.4787
1.8974	1.6288	1600	0.5974	0.4545
1.7759	1.8324	1800	0.5868	0.4709
1.0256	2.0366	2000	0.5886	0.4393
1.2131	2.2401	2200	0.6019	0.4494
0.996	2.4437	2400	0.5997	0.4421

Framework versions

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