Speed up time between trials?

[apal0934]

I'm new to ML, and Autokeras is working great for me so far with the StructuredDataClassifer, except for one issue. The time during training is quite fast, but the time between trials is unbearably slow. My input is a 640kb, 3k lines CSV with 60 columns (blows out to thousands of columns with one hot encoding), and I have a 3080TI and Ryzen 9 5950x with CUDA properly set up.

I tried looking into profiling tools like the TF Profiler, but it only profiled the steps. I've also tried setting various env vars such as CUDA_CACHE_MAXSIZE and TF_GPU_THREAD_MODE, but to no avail.

Is there anything else I could be looking at to speed up the time for a trial to start?