RuntimeError: Can't compile: Cannot find crypto parameters

[bpradeep508]

n_bits = 6

q_module = compile_torch_model(model, X_train, rounding_threshold_bits=6, p_error=0.1)

start_time = time.time()
accs = test_with_concrete(
q_module,
test_dataloader,
use_sim=True,
)
sim_time = time.time() - start_time

print(f"Simulated FHE execution for {n_bits} bit network accuracy: {accs:.2f}%")

error from second line. here X_train is not a mnist dataset given in example, custom dataset

[jfrery]

Hi @bpradeep508,

Looks like this is the code from the CNN example. Have you changed the data or model in the example?

Higher p_error, or a adding the n_bits = 6 might help. But to understand why you see this error precisely we would need the architecture of your model.