Hanhpt23/whisper-medium-Encode-GermanMed-full

openai/whisper-medium

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

• Loss: 0.8743
• Wer: 26.2573

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
0.7043	1.0	194	0.7433	43.3508
0.4383	2.0	388	0.7578	37.6118
0.28	3.0	582	0.8223	39.6380
0.188	4.0	776	0.8428	35.1538
0.1479	5.0	970	0.8755	32.6751
0.1263	6.0	1164	0.8562	31.2249
0.0808	7.0	1358	0.8797	31.5129
0.063	8.0	1552	0.9294	33.3333
0.0469	9.0	1746	0.9285	35.4315
0.0464	10.0	1940	0.9110	29.5176
0.0302	11.0	2134	0.9158	33.4568
0.0355	12.0	2328	0.9420	31.9243
0.0167	13.0	2522	0.9098	30.6284
0.0119	14.0	2716	0.8894	29.7645
0.0092	15.0	2910	0.8861	26.9567
0.0034	16.0	3104	0.8764	26.9670
0.0007	17.0	3298	0.8692	26.2573
0.0007	18.0	3492	0.8724	26.6584
0.0002	19.0	3686	0.8739	26.2265
0.0002	20.0	3880	0.8743	26.2573

Framework versions

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