whisper-medium-bsbigcgen-combined-model

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

• Loss: 0.5866
• Wer: 0.4674

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: 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
• num_epochs: 30.0
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.2954	0.3146	200	0.9741	0.6849
3.2141	0.6292	400	0.7601	0.5826
2.7215	0.9438	600	0.6713	0.4974
2.1639	1.2595	800	0.6447	0.5168
1.9576	1.5741	1000	0.6139	0.4731
2.0274	1.8887	1200	0.5895	0.4721
1.1246	2.2045	1400	0.5959	0.4556
1.1317	2.5191	1600	0.6155	0.4656
1.0119	2.8337	1800	0.5866	0.4674
0.7011	3.1494	2000	0.6415	0.4684
0.5972	3.4640	2200	0.6469	0.4687
0.6218	3.7786	2400	0.6497	0.4226

Framework versions

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