diff --git a/src/metatrain/experimental/nanopet/tests/test_regression.py b/src/metatrain/experimental/nanopet/tests/test_regression.py
index c4c0aa3f9..ad3f19eb9 100644
--- a/src/metatrain/experimental/nanopet/tests/test_regression.py
+++ b/src/metatrain/experimental/nanopet/tests/test_regression.py
@@ -115,11 +115,11 @@ def test_regression_train():
 
     expected_output = torch.tensor(
         [
-            [-0.162086308002],
-            [-0.022639814764],
-            [0.000784173608],
-            [0.019549306482],
-            [0.063824191689],
+            [0.99948340654373168945],
+            [0.58770644664764404297],
+            [0.26674023270606994629],
+            [0.53543293476104736328],
+            [0.25562191009521484375],
         ]
     )
 
diff --git a/tests/utils/test_llpr.py b/tests/utils/test_llpr.py
index 46e98f21e..a6be4583e 100644
--- a/tests/utils/test_llpr.py
+++ b/tests/utils/test_llpr.py
@@ -102,7 +102,7 @@ def test_llpr(tmpdir):
             params.append(param.squeeze())
     weights = torch.cat(params)
 
-    n_ensemble_members = 10000
+    n_ensemble_members = 1000000  # converges slowly...
     llpr_model.calibrate(dataloader)
     llpr_model.generate_ensemble({"energy": weights}, n_ensemble_members)
     assert "energy_ensemble" in llpr_model.capabilities.outputs
@@ -143,7 +143,7 @@ def test_llpr(tmpdir):
     )
 
     torch.testing.assert_close(
-        analytical_uncertainty, ensemble_uncertainty, rtol=1e-2, atol=1e-2
+        analytical_uncertainty, ensemble_uncertainty, rtol=5e-3, atol=0.0
     )