The model performs poorly under the same conditions

[deeomnjobs]

I used sklearn.svc and thundersvm.svc respectively, thundersvm was very fast to train compared to sklearn.svc, but the final training results were poor, sklearn.svc ended up with an acc of about 0.7 while thundersvm only had about 0.5.
I used the default parameters.svm.SVC() and thundersvm.SVC()

[Shentl]

Same problem, in a 20-class classification problem, sklearn.svc achieves an acc of 0.87, but thundersvm.SVC() only achieves an acc of 0.057. And the output class of thundersvm is limited to 4 while there are 20 classes in the dataset