whisper-hakka-small-1

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

• Loss: 0.6134
• Cer: 839.6261

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: 8
• eval_batch_size: 8
• seed: 42
• gradient_accumulation_steps: 2
• 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: 500
• training_steps: 20000

Training results

Training Loss	Epoch	Step	Validation Loss	Cer
0.1701	0.78	1000	0.3828	52.2939
0.0764	1.55	2000	0.3519	36.2758
0.0277	2.33	3000	0.3274	40.2633
0.013	3.1	4000	0.4307	112.1952
0.0092	3.88	5000	0.4454	37.3418
0.007	4.66	6000	0.4402	70.9902
0.0036	5.43	7000	0.4313	77.5403
0.0035	6.21	8000	0.5371	104.9887
0.0038	6.98	9000	0.5463	185.2789
0.0018	7.76	10000	0.5453	184.7941
0.0011	8.54	11000	0.5113	213.0442
0.0012	9.31	12000	0.6346	402.5504
0.0005	10.09	13000	0.5790	399.8495
0.0002	10.87	14000	0.5790	491.2496
0.0003	11.64	15000	0.6038	652.7718
0.0001	12.42	16000	0.6070	814.4883
0.0001	13.19	17000	0.6078	801.3233
0.0001	13.97	18000	0.6068	772.7599
0.0001	14.75	19000	0.6160	802.9302
0.0001	15.52	20000	0.6134	839.6261

Framework versions

• Transformers 4.36.0
• Pytorch 2.0.1
• Datasets 2.14.6
• Tokenizers 0.15.0