whisper-base-zh-20230718-1 - au2a

This model is a fine-tuned version of openai/whisper-small on the some hakka audio dataset. It achieves the following results on the evaluation set:

• Loss: 0.4142
• Cer: 84.7926

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-06
• train_batch_size: 64
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• total_train_batch_size: 128
• optimizer: Adam with betas=(0.9,0.999) and epsilon=1e-08
• lr_scheduler_type: linear
• lr_scheduler_warmup_steps: 500
• training_steps: 10000

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.0499	2.59	1000	0.3377	153.9019
0.0035	5.17	2000	0.3506	138.4528
0.0015	7.76	3000	0.3651	128.2541
0.001	10.35	4000	0.3754	105.1522
0.0005	12.94	5000	0.3841	90.0846
0.0004	15.52	6000	0.3925	92.5134
0.0002	18.11	7000	0.4011	86.3035
0.0002	20.7	8000	0.4070	80.0219
0.0001	23.29	9000	0.4118	82.5451
0.0001	25.87	10000	0.4142	84.7926

Framework versions

• Transformers 4.30.2
• Pytorch 2.0.1+cu117
• Datasets 2.13.1
• Tokenizers 0.13.3