Hanhpt23/whisper-medium-GermanMed-v1

openai/whisper-medium

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

• Loss: 5.5611
• Wer: 107.4053

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
3.0873	1.0	765	3.7193	114.9828
2.0404	2.0	1530	3.8002	105.8553
1.1104	3.0	2295	4.2393	106.7738
0.6064	4.0	3060	4.4641	113.7773
0.3464	5.0	3825	4.8067	106.8886
0.2769	6.0	4590	5.0650	106.6590
0.2327	7.0	5355	5.0613	111.1366
0.181	8.0	6120	5.2019	104.7072
0.17	9.0	6885	5.3335	123.8806
0.1518	10.0	7650	5.3649	156.4294
0.1405	11.0	8415	5.4181	107.6349
0.1229	12.0	9180	5.4629	102.0666
0.1335	13.0	9945	5.4842	106.0850
0.1016	14.0	10710	5.4736	105.4535
0.119	15.0	11475	5.4178	109.1848
0.0979	16.0	12240	5.4872	106.7738
0.0919	17.0	13005	5.5066	105.7979
0.0835	18.0	13770	5.5156	117.8530
0.0979	19.0	14535	5.5140	109.9885
0.1001	20.0	15300	5.5611	107.4053

Framework versions

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