./whisper-large-cit-synth-do0.15-wd0-lr1e-05-1000

This model is a fine-tuned version of openai/whisper-large-v3 on the SF 1000 dataset. It achieves the following results on the evaluation set:

• Loss: 0.4507
• Wer: 21.8713

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: 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: 100
• training_steps: 300
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.6733	0.3556	20	0.4937	29.2008
0.4835	0.7111	40	0.3850	24.2105
0.4129	1.0667	60	0.3589	23.5088
0.268	1.4222	80	0.3472	23.4698
0.2525	1.7778	100	0.3474	23.0019
0.1903	2.1333	120	0.3608	22.7680
0.1316	2.4889	140	0.3730	22.9240
0.1368	2.8444	160	0.3545	25.3801
0.088	3.2	180	0.3879	23.0409
0.0688	3.5556	200	0.4038	23.9376
0.0672	3.9111	220	0.3813	22.1832
0.0449	4.2667	240	0.4250	22.8070
0.0338	4.6222	260	0.4314	22.2222
0.0376	4.9778	280	0.4250	21.4425
0.0183	5.3333	300	0.4507	21.8713

Framework versions

• Transformers 4.42.3
• Pytorch 1.13.1+cu117
• Datasets 2.20.0
• Tokenizers 0.19.1