ORF estimate takes too long

I am using the DiscreteTreatmentOrthoForest's const_marginal_effect() method to estimate my test data, which is 4840 by 24. I used a trained model and call model.const_marginal_effect(test_data) directly. It takes hours to estimated the effect. I am wondering if I am doing correctly or there is any way to speed up the process? Here is my code for reference. Thank you so much!

#Load trained model
model = pickle.load(open('ModelPath', 'rb'))

#Load test data, shape (4840, 24)
test_data = pickle.load(open('TestDataPath', 'rb')).values

#Estimate the treatment effect
TE = model.const_marginal_effect(test_data)

And here is the first few lines of terminal prints:
[Parallel(n_jobs=-1)]: Using backend ThreadingBackend with 4 concurrent workers.
[Parallel(n_jobs=-1)]: Done 24 tasks | elapsed: 9.9min
[Parallel(n_jobs=-1)]: Done 120 tasks | elapsed: 54.4min

Start with DiscreteTreatmentOrthoForest.const_marginal_effect() and the supplied 4,840×24 test-data example, then review the terminal's Parallel timing output. Done should establish whether the call is being used correctly and identify a supported way to reduce the runtime, if one exists.