diff --git a/src/bioemu/model_utils.py b/src/bioemu/model_utils.py
index d70d513..53a8626 100644
--- a/src/bioemu/model_utils.py
+++ b/src/bioemu/model_utils.py
@@ -41,13 +41,13 @@ def maybe_download_checkpoint(
 
     except HTTPError as e:
         fs = HfFileSystem()
-        available_checkpoints = [
+        available_checkpoints = {
             Path(p).parent.name for p in fs.glob("microsoft/bioemu/checkpoints/*/checkpoint.ckpt")
-        ]
+        }
         available_configs = [
             Path(p).parent.name for p in fs.glob("microsoft/bioemu/checkpoints/*/config.yaml")
         ]
-        available_model_names = sorted(set(available_checkpoints).intersection(available_configs))
+        available_model_names = sorted(available_checkpoints.intersection(available_configs))
         raise ValueError(
             f"Model {model_name} not found. Available model names: " f"{available_model_names}"
         ) from e
diff --git a/src/bioemu/sample.py b/src/bioemu/sample.py
index 1ec9f50..2a5ed8b 100644
--- a/src/bioemu/sample.py
+++ b/src/bioemu/sample.py
@@ -193,9 +193,9 @@ def main(
 
     logger.info("Converting samples to .pdb and .xtc...")
     samples_files = sorted(list(output_dir.glob("batch_*.npz")))
-    sequences = [np.load(f)["sequence"].item() for f in samples_files]
-    if set(sequences) != {sequence}:
-        raise ValueError(f"Expected all sequences to be {sequence}, but got {set(sequences)}")
+    sequences = {np.load(f)["sequence"].item() for f in samples_files}
+    if sequences != {sequence}:
+        raise ValueError(f"Expected all sequences to be {sequence}, but got {sequences}")
     positions = torch.tensor(np.concatenate([np.load(f)["pos"] for f in samples_files]))
     node_orientations = torch.tensor(
         np.concatenate([np.load(f)["node_orientations"] for f in samples_files])
diff --git a/src/bioemu/training/foldedness.py b/src/bioemu/training/foldedness.py
index c8e4126..c0ce257 100644
--- a/src/bioemu/training/foldedness.py
+++ b/src/bioemu/training/foldedness.py
@@ -87,8 +87,8 @@ def compute_fnc_for_list(batch: list[ChemGraph], reference_info: ReferenceInfo)
     Returns:
         torch tensor of fraction of native contacts.
     """
-    seqs = [x.sequence for x in batch]
-    assert len(set(seqs)) == 1, "Batch should contain samples all from the same system."
+    seqs = {x.sequence for x in batch}
+    assert len(seqs) == 1, "Batch should contain samples all from the same system."
     sequence = seqs[0]
 
     device = batch[0].pos.device
diff --git a/src/bioemu/training/loss.py b/src/bioemu/training/loss.py
index e972e3c..b7800cc 100644
--- a/src/bioemu/training/loss.py
+++ b/src/bioemu/training/loss.py
@@ -145,8 +145,8 @@ def _estimate_squared_mean_error(
         loss: an estimate of [(mean foldedness of samples) - (target mean foldedness)]^2.
     """
     assert isinstance(batch, list)  # Not a Batch!
-    sequences = [x.sequence for x in batch]
-    assert len(set(sequences)) == 1, "Batch must contain samples all from the same system."
+    sequences = {x.sequence for x in batch}
+    assert len(sequences) == 1, "Batch must contain samples all from the same system."
     n = len(batch)
     assert n >= 2, "Batch must contain at least two samples."