Hanhpt23/whisper-base-GermanMed-full

openai/whisper-base

This model is a fine-tuned version of openai/whisper-base on the Hanhpt23/GermanMed-full dataset. It achieves the following results on the evaluation set:

• Loss: 0.7401
• Wer: 24.4575

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: 0.0001
• train_batch_size: 8
• eval_batch_size: 8
• seed: 42
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 100
• num_epochs: 20

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.6052	1.0	194	0.5948	34.0636
0.2976	2.0	388	0.5938	29.7336
0.1442	3.0	582	0.6085	34.0327
0.0783	4.0	776	0.6296	26.1339
0.055	5.0	970	0.6500	27.4812
0.0359	6.0	1164	0.6595	28.2012
0.0263	7.0	1358	0.7021	27.1418
0.0218	8.0	1552	0.7004	28.2217
0.0151	9.0	1746	0.7171	28.5303
0.0136	10.0	1940	0.7387	25.8665
0.0108	11.0	2134	0.7312	25.5580
0.0042	12.0	2328	0.7272	26.3910
0.0017	13.0	2522	0.7298	25.3625
0.0028	14.0	2716	0.7286	24.6015
0.0004	15.0	2910	0.7271	24.7557
0.0004	16.0	3104	0.7362	24.6940
0.0004	17.0	3298	0.7358	24.3855
0.0005	18.0	3492	0.7383	24.2723
0.0003	19.0	3686	0.7397	24.4369
0.0003	20.0	3880	0.7401	24.4575

Framework versions

• Transformers 4.41.1
• Pytorch 2.3.0
• Datasets 2.19.1
• Tokenizers 0.19.1