whisper-small-zh-20230714-1 - 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: 0.3765
• Cer: 85.1930

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
• 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
0.1072	1.29	1000	0.3484	60.5235
0.0176	2.59	2000	0.3341	58.3327
0.0031	3.88	3000	0.3407	76.3806
0.0015	5.17	4000	0.3496	72.6222
0.001	6.47	5000	0.3591	64.4028
0.0012	7.76	6000	0.3598	51.4316
0.0005	9.06	7000	0.3663	69.3171
0.0004	10.35	8000	0.3696	81.2420
0.0003	11.64	9000	0.3746	84.8833
0.0003	12.94	10000	0.3765	85.1930

Framework versions

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