Hanhpt23/whisper-small-Encode-GermanMed-full

openai/whisper-small

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

• Loss: 0.7961
• Wer: 26.1648

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
0.6649	1.0	194	0.6657	43.7108
0.3613	2.0	388	0.6721	38.5375
0.2014	3.0	582	0.6927	38.8769
0.1383	4.0	776	0.7546	35.0098
0.1053	5.0	970	0.7698	34.0636
0.086	6.0	1164	0.7729	29.7028
0.059	7.0	1358	0.7985	36.8405
0.0471	8.0	1552	0.8244	30.3919
0.039	9.0	1746	0.8291	30.2067
0.0195	10.0	1940	0.8342	33.1379
0.0149	11.0	2134	0.8184	30.7004
0.0103	12.0	2328	0.8249	29.4868
0.0077	13.0	2522	0.8106	33.0351
0.0039	14.0	2716	0.7991	29.0445
0.0017	15.0	2910	0.8102	28.0160
0.0019	16.0	3104	0.7934	26.5247
0.0014	17.0	3298	0.7996	26.7201
0.0002	18.0	3492	0.7955	26.5659
0.0002	19.0	3686	0.7959	26.2059
0.0003	20.0	3880	0.7961	26.1648

Framework versions

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