whisper-medium-bigcgen-combined-30hrs-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: 0.4850
• Wer: 0.3936

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
3.9757	0.1016	200	0.8962	0.6582
3.5816	0.2031	400	0.7813	0.6241
3.0158	0.3047	600	0.7028	0.5563
2.8825	0.4063	800	0.6599	0.5168
2.2129	0.5079	1000	0.6258	0.5043
2.6478	0.6094	1200	0.6042	0.4907
2.4429	0.7110	1400	0.5742	0.4566
2.3049	0.8126	1600	0.5661	0.4520
2.3745	0.9142	1800	0.5507	0.4601
2.0113	1.0157	2000	0.5367	0.4423
1.8125	1.1173	2200	0.5371	0.4177
1.4508	1.2189	2400	0.5415	0.4146
1.5437	1.3205	2600	0.5249	0.4325
1.4371	1.4220	2800	0.5204	0.4072
1.5153	1.5236	3000	0.5091	0.3927
1.4276	1.6252	3200	0.5051	0.4051
1.6076	1.7268	3400	0.4975	0.3990
1.5648	1.8283	3600	0.4922	0.3949
1.4718	1.9299	3800	0.4850	0.3936
0.9453	2.0315	4000	0.5069	0.3853
0.7122	2.1331	4200	0.5143	0.3853
1.0139	2.2346	4400	0.5074	0.3919

Framework versions

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