diff --git a/tests/models/test_encoder_processor_decoder.py b/tests/models/test_encoder_processor_decoder.py
deleted file mode 100644
index 7ded836..0000000
--- a/tests/models/test_encoder_processor_decoder.py
+++ /dev/null
@@ -1,49 +0,0 @@
-from pathlib import Path
-
-import pytest
-import torch
-
-from anemoi.models.models.encoder_processor_decoder import AnemoiModelEncProcDec
-
-
-@pytest.mark.gpu()
-@pytest.mark.parametrize(
-    "config",
-    [
-        [],
-        ["data.forcing=[]"],
-        ["data.forcing=null"],
-        ["data.forcing=[lsm]"],
-        ["data.diagnostic=[]"],
-        ["data.diagnostic=null"],
-        ["data.diagnostic=[msl]"],
-        ["data.forcing=null", "data.diagnostic=null"],
-        ["data.forcing=[]", "data.diagnostic=[]"],
-        ["data.forcing=[lsm]", "data.diagnostic=[]"],
-        ["data.forcing=[]", "data.diagnostic=[lsm]"],
-        ["data.forcing=[lsm]", "data.diagnostic=[tp]"],
-    ],
-    indirect=True,
-)
-def test_graph_msg(config, datamodule) -> None:
-    device = torch.device("cuda")
-    data_indices = datamodule.data_indices
-    config.data.num_features = len(datamodule.ds_train.data.variables)
-    graph_data = torch.load(Path(config.hardware.paths.graph, config.hardware.files.graph))
-    model = AnemoiModelEncProcDec(
-        config=config,
-        data_indices=data_indices,
-        graph_data=graph_data,
-    ).to(device)
-
-    input_data = torch.randn(
-        config.dataloader.batch_size.training,
-        config.training.multistep_input,
-        1,
-        40320,
-        len(data_indices.model.input.full),
-        dtype=torch.float32,
-        device=device,
-    )
-    output = torch.randn([config.dataloader.batch_size.training, 1, 40320, len(data_indices.model.output.full)])
-    assert model.forward(input_data).shape == output.shape, "Output shape is not correct"
diff --git a/tests/test_models.py b/tests/models/test_models.py
similarity index 100%
rename from tests/test_models.py
rename to tests/models/test_models.py