Hanhpt23/whisper-tiny-Encode-GermanMed-full

openai/whisper-tiny

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

• Loss: 0.9461
• Wer: 30.9061

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
0.8146	1.0	194	0.8137	65.1856
0.4421	2.0	388	0.8220	37.1285
0.2251	3.0	582	0.7980	39.5557
0.1636	4.0	776	0.8563	50.7457
0.0827	5.0	970	0.8480	40.8516
0.0772	6.0	1164	0.8860	43.8136
0.0437	7.0	1358	0.9120	37.8793
0.0328	8.0	1552	0.9252	34.8144
0.0222	9.0	1746	0.9330	35.4520
0.0216	10.0	1940	0.9464	33.9504
0.0145	11.0	2134	0.9413	32.3151
0.0072	12.0	2328	0.9746	33.8990
0.0045	13.0	2522	0.9515	32.3871
0.0024	14.0	2716	0.9588	34.3618
0.0031	15.0	2910	0.9483	34.0533
0.0006	16.0	3104	0.9485	30.8135
0.0005	17.0	3298	0.9433	30.8444
0.0004	18.0	3492	0.9449	31.0398
0.0004	19.0	3686	0.9457	30.9575
0.0004	20.0	3880	0.9461	30.9061

Framework versions

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