Mezosky/whisper-large-v3-clinical-assistance

Whisper Chilean Spanish Large v3

This model is a fine-tuned version of openai/whisper-large-v3 on the Mezosky/es_clinical_assistance_10k dataset. It achieves the following results on the evaluation set:

• Loss: 0.0961
• Wer: 6.9352

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: 1e-05
• train_batch_size: 16
• 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: 500
• training_steps: 2000

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.2816	0.17	100	0.2250	11.2827
0.1505	0.34	200	0.1479	9.8196
0.1293	0.51	300	0.1350	72.1192
0.1221	0.69	400	0.1292	9.6825
0.141	0.86	500	0.1194	53.0899
0.0922	1.03	600	0.1150	12.0380
0.0773	1.2	700	0.1079	12.8661
0.0745	1.37	800	0.1036	67.3017
0.0699	1.54	900	0.1016	8.2697
0.0917	1.72	1000	0.0956	8.6334
0.0716	1.89	1100	0.0968	7.7997
0.0441	2.06	1200	0.0946	8.3760
0.0377	2.23	1300	0.0963	7.6178
0.0417	2.4	1400	0.0951	7.5703
0.0409	2.57	1500	0.0926	7.2681
0.0356	2.74	1600	0.0912	6.8933
0.0361	2.92	1700	0.0918	7.0835
0.0215	3.09	1800	0.0938	6.9548
0.018	3.26	1900	0.0960	6.6415
0.0196	3.43	2000	0.0961	6.9352

Framework versions

• Transformers 4.39.3
• Pytorch 2.2.2+cu121
• Datasets 2.18.0
• Tokenizers 0.15.2