Hanhpt23/whisper-base-Encode-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: 0.8154
• Wer: 28.0469

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
0.694	1.0	194	0.6826	38.7329
0.3399	2.0	388	0.6960	39.2780
0.1866	3.0	582	0.7193	42.8057
0.1157	4.0	776	0.7415	31.9963
0.0853	5.0	970	0.7792	32.7265
0.0574	6.0	1164	0.7686	33.7036
0.0358	7.0	1358	0.7852	33.1791
0.0423	8.0	1552	0.8025	32.1917
0.027	9.0	1746	0.8138	31.2044
0.0172	10.0	1940	0.8155	29.0548
0.0182	11.0	2134	0.8280	30.7004
0.0053	12.0	2328	0.8268	30.4227
0.0066	13.0	2522	0.8221	30.0319
0.0023	14.0	2716	0.8166	29.5176
0.0007	15.0	2910	0.8215	28.6948
0.0012	16.0	3104	0.8262	28.4789
0.0003	17.0	3298	0.8158	28.3143
0.0004	18.0	3492	0.8154	27.9338
0.0003	19.0	3686	0.8154	27.9646
0.0003	20.0	3880	0.8154	28.0469

Framework versions

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