CheeLi03/whisper-5b-20k

Whisper Small Five 20K - Chee Li

This model is a fine-tuned version of openai/whisper-base on the Google Fleurs dataset. It achieves the following results on the evaluation set:

• Loss: 0.5771
• Wer: 22.0375

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: 2500
• training_steps: 20000
• mixed_precision_training: Native AMP

Training results

Training Loss	Epoch	Step	Validation Loss	Wer
0.4014	1.0560	1000	0.4369	25.7071
0.2677	2.1119	2000	0.3905	22.1327
0.1651	3.1679	3000	0.3856	21.2139
0.1102	4.2239	4000	0.3920	20.4471
0.0514	5.2798	5000	0.4072	21.2883
0.0255	6.3358	6000	0.4273	21.4687
0.0184	7.3918	7000	0.4442	21.6251
0.01	8.4477	8000	0.4635	21.3397
0.0051	9.5037	9000	0.4805	21.3867
0.0043	10.5597	10000	0.4924	21.5508
0.0025	11.6156	11000	0.5054	21.5847
0.0023	12.6716	12000	0.5166	22.0703
0.0016	13.7276	13000	0.5292	21.7509
0.0012	14.7835	14000	0.5375	21.7925
0.001	15.8395	15000	0.5480	21.9325
0.0008	16.8955	16000	0.5565	21.8866
0.0008	17.9514	17000	0.5638	21.9423
0.0005	19.0074	18000	0.5709	21.9916
0.0005	20.0634	19000	0.5755	22.0397
0.0004	21.1193	20000	0.5771	22.0375

Framework versions

• Transformers 4.43.4
• Pytorch 2.3.1+cu121
• Datasets 2.20.0
• Tokenizers 0.19.1