skip cycle consistency test for diffusion models

- the test is unstable for untrained diffusion models, as the networks
  output is not sufficiently smooth for the step size we use
- remove the diffusion_model marker

Author: vpratz

tests/test_networks/conftest.py

@@ -158,12 +158,12 @@ def typical_point_inference_network_subnet():
         "flow_matching",
         "free_form_flow",
         "consistency_model",
-        pytest.param("diffusion_model_edm_F", marks=pytest.mark.diffusion_model),
-        pytest.param("diffusion_model_edm_noise", marks=[pytest.mark.slow, pytest.mark.diffusion_model]),
-        pytest.param("diffusion_model_cosine_velocity", marks=[pytest.mark.slow, pytest.mark.diffusion_model]),
-        pytest.param("diffusion_model_cosine_F", marks=[pytest.mark.slow, pytest.mark.diffusion_model]),
-        pytest.param("diffusion_model_cosine_noise", marks=[pytest.mark.slow, pytest.mark.diffusion_model]),
-        pytest.param("diffusion_model_cosine_velocity", marks=[pytest.mark.slow, pytest.mark.diffusion_model]),
+        pytest.param("diffusion_model_edm_F"),
+        pytest.param("diffusion_model_edm_noise", marks=pytest.mark.slow),
+        pytest.param("diffusion_model_cosine_velocity", marks=pytest.mark.slow),
+        pytest.param("diffusion_model_cosine_F", marks=pytest.mark.slow),
+        pytest.param("diffusion_model_cosine_noise", marks=pytest.mark.slow),
+        pytest.param("diffusion_model_cosine_velocity", marks=pytest.mark.slow),
     ],
     scope="function",
 )
@@ -191,37 +191,33 @@ def inference_network_subnet(request):
         "flow_matching",
         "free_form_flow",
         "consistency_model",
-        pytest.param("diffusion_model_edm_F", marks=pytest.mark.diffusion_model),
+        pytest.param("diffusion_model_edm_F"),
         pytest.param(
             "diffusion_model_edm_noise",
             marks=[
                 pytest.mark.slow,
-                pytest.mark.diffusion_model,
                 pytest.mark.skip("noise predicition not testable without prior training for numerical reasons."),
             ],
         ),
-        pytest.param("diffusion_model_cosine_velocity", marks=[pytest.mark.slow, pytest.mark.diffusion_model]),
+        pytest.param("diffusion_model_cosine_velocity", marks=pytest.mark.slow),
         pytest.param(
             "diffusion_model_cosine_F",
             marks=[
                 pytest.mark.slow,
-                pytest.mark.diffusion_model,
                 pytest.mark.skip("skip to reduce load on CI."),
             ],
         ),
         pytest.param(
             "diffusion_model_cosine_noise",
             marks=[
                 pytest.mark.slow,
-                pytest.mark.diffusion_model,
                 pytest.mark.skip("noise predicition not testable without prior training for numerical reasons."),
             ],
         ),
         pytest.param(
             "diffusion_model_cosine_velocity",
             marks=[
                 pytest.mark.slow,
-                pytest.mark.diffusion_model,
                 pytest.mark.skip("skip to reduce load on CI."),
             ],
         ),

tests/test_networks/test_inference_networks.py

@@ -86,13 +86,16 @@ def test_output_shape(generative_inference_network, random_samples, random_condi
 
 def test_cycle_consistency(generative_inference_network, random_samples, random_conditions):
     # cycle-consistency means the forward and inverse methods are inverses of each other
+    import bayesflow as bf
+
+    if isinstance(generative_inference_network, bf.experimental.DiffusionModel):
+        pytest.skip(reason="test unstable for untrained diffusion models")
     try:
         forward_output, forward_log_density = generative_inference_network(
             random_samples, conditions=random_conditions, density=True
         )
     except NotImplementedError:
-        # network is not invertible, cycle consistency cannot be tested.
-        return
+        pytest.skip(reason="network is not invertible")
     inverse_output, inverse_log_density = generative_inference_network(
         forward_output, conditions=random_conditions, density=True, inverse=True
     )