Hanhpt23/whisper-base-GermanMed-v1

openai/whisper-base

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

• Loss: 5.5096
• Wer: 105.0517

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
3.094	1.0	765	3.6319	147.1871
2.2494	2.0	1530	3.7314	101.2055
1.5508	3.0	2295	3.9256	109.4719
0.9752	4.0	3060	4.2097	106.3720
0.4449	5.0	3825	4.5738	103.4443
0.2699	6.0	4590	4.7393	113.6625
0.2048	7.0	5355	4.9283	105.8553
0.1749	8.0	6120	4.9945	110.6200
0.1707	9.0	6885	5.1057	121.0677
0.1193	10.0	7650	5.1520	116.5327
0.1266	11.0	8415	5.3014	107.0608
0.1156	12.0	9180	5.3254	104.9369
0.1282	13.0	9945	5.3345	123.4214
0.0944	14.0	10710	5.3766	129.9656
0.113	15.0	11475	5.3981	109.5293
0.0986	16.0	12240	5.4435	104.5350
0.0919	17.0	13005	5.4113	105.9701
0.0824	18.0	13770	5.4515	102.2962
0.0974	19.0	14535	5.4659	108.0367
0.0995	20.0	15300	5.5096	105.0517

Framework versions

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