Makkoen/whisper-medium.en-cit-do015-wd0-lr3e-06-FULL3

./3382

This model is a fine-tuned version of openai/whisper-medium.en on the 3382 NYC 1000 dataset. It achieves the following results on the evaluation set:

• Loss: 0.6304
• Wer Ortho: 32.2501
• Wer: 23.5222

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: 3e-06
• train_batch_size: 4
• eval_batch_size: 8
• seed: 42
• distributed_type: multi-GPU
• gradient_accumulation_steps: 4
• total_train_batch_size: 16
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 200
• training_steps: 1000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer Ortho	Wer
1.5524	0.5256	100	1.0430	42.1375	33.6570
1.0535	1.0512	200	0.8779	37.1493	27.9815
0.8222	1.5769	300	0.7495	35.4208	26.5674
0.6909	2.1025	400	0.6826	33.2082	24.5121
0.5843	2.6281	500	0.6558	32.8625	24.1350
0.5347	3.1537	600	0.6436	32.4773	23.5693
0.4819	3.6794	700	0.6377	33.5243	24.4555
0.4922	4.2050	800	0.6338	31.9933	23.0980
0.4638	4.7306	900	0.6318	32.1513	23.4845
0.4362	5.2562	1000	0.6304	32.2501	23.5222

Framework versions

• Transformers 4.44.0
• Pytorch 1.13.1+cu117
• Datasets 2.21.0
• Tokenizers 0.19.1