saurabhy27-outcomes/whisper-large-v3-medical

Whisper Large V3 Medical

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

• Loss: 0.1453
• Wer: 3.2636

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: 5e-07
• train_batch_size: 64
• eval_batch_size: 32
• seed: 42
• distributed_type: multi-GPU
• optimizer: Use adamw_torch with betas=(0.9,0.999) and epsilon=1e-08 and optimizer_args=No additional optimizer arguments
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• training_steps: 5000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
4.2439	0.1530	200	0.2935	4.5078
3.3374	0.3060	400	0.2734	4.6961
3.0833	0.4591	600	0.2673	4.2733
1.8243	0.6121	800	0.2681	4.4373
1.1288	0.7651	1000	0.2549	4.2771
0.8199	0.9181	1200	0.2412	4.2041
0.681	1.0712	1400	0.2311	4.1054
0.5798	1.2242	1600	0.2192	4.0093
0.5233	1.3772	1800	0.2072	3.8927
0.463	1.5302	2000	0.1992	3.8197
0.428	1.6832	2200	0.1951	3.7748
0.3944	1.8363	2400	0.1866	3.6775
0.3682	1.9893	2600	0.1792	3.6044
0.3543	2.1423	2800	0.1725	3.5301
0.3368	2.2953	3000	0.1714	3.4904
0.3136	2.4484	3200	0.1648	3.4571
0.3121	2.6014	3400	0.1604	3.4238
0.2959	2.7544	3600	0.1561	3.3956
0.2912	2.9074	3800	0.1538	3.3738
0.2767	3.0604	4000	0.1511	3.3456
0.2848	3.2135	4200	0.1487	3.3200
0.274	3.3665	4400	0.1475	3.2841
0.2694	3.5195	4600	0.1464	3.2828
0.2731	3.6725	4800	0.1455	3.2687
0.2677	3.8256	5000	0.1453	3.2636

Framework versions

• Transformers 4.48.0.dev0
• Pytorch 2.1.0+cu118
• Datasets 3.1.1.dev0
• Tokenizers 0.21.0