fix: hybrid test

tests/torch/test_hybrid_converter.py

@@ -329,9 +329,9 @@ def prepare_data(x, y, test_size=0.1, random_state=42):
     y_glwe = hybrid_local(x1_test, fhe="execute").numpy()
 
     y1_test = y1_test.numpy()
-    acc_fp32 = numpy.sum(numpy.argmax(y_torch, axis=1) == y1_test)
-    acc_qm = numpy.sum(numpy.argmax(y_qm, axis=1) == y1_test)
-    acc_glwe = numpy.sum(numpy.argmax(y_glwe, axis=1) == y1_test)
+    n_correct_fp32 = numpy.sum(numpy.argmax(y_torch, axis=1) == y1_test)
+    n_correct_qm = numpy.sum(numpy.argmax(y_qm, axis=1) == y1_test)
+    n_correct_glwe = numpy.sum(numpy.argmax(y_glwe, axis=1) == y1_test)
 
     # These two should be exactly the same
     assert numpy.all(numpy.allclose(y_torch, y_hybrid_torch, rtol=1, atol=0.001))
@@ -349,9 +349,9 @@ def prepare_data(x, y, test_size=0.1, random_state=42):
     assert numpy.all(numpy.allclose(y_qm, y_glwe, rtol=1, atol=threshold_fhe))
     assert numpy.all(numpy.allclose(y_torch, y_glwe, rtol=1, atol=threshold_fhe))
 
-    acc_threshold_fhe = 0.01
+    n_correct_delta_threshold_fhe = 1
     # Check accuracy between fp32 and glwe
-    assert numpy.abs(acc_fp32 - acc_glwe) < acc_threshold_fhe
+    assert numpy.abs(n_correct_fp32 - n_correct_glwe) <= n_correct_delta_threshold_fhe
 
     # Check accuracy between quantized and glwe
-    assert numpy.abs(acc_qm - acc_glwe) < acc_threshold_fhe
+    assert numpy.abs(n_correct_qm - n_correct_glwe) <= n_correct_delta_threshold_fhe