diff --git a/src/plaid/infos.py b/src/plaid/infos.py
index a5180295..159ab668 100644
--- a/src/plaid/infos.py
+++ b/src/plaid/infos.py
@@ -41,6 +41,22 @@ class DataProduction(
 )
 
 
+def _num_samples_sort_key(key: str) -> tuple[int, str]:
+    """Return the serialization sort key for ``num_samples`` split names."""
+    if key.startswith("train"):
+        return (0, key)
+    if key.startswith("test"):
+        return (1, key)
+    return (2, key)
+
+
+def _sort_num_samples_keys(num_samples: dict[str, int]) -> dict[str, int]:
+    """Sort ``num_samples`` with train* keys first, then test*, then others."""
+    return {
+        key: num_samples[key] for key in sorted(num_samples, key=_num_samples_sort_key)
+    }
+
+
 class Infos(
     BaseModel,
     revalidate_instances="always",
@@ -205,6 +221,8 @@ def save_to_file(self, path: Union[str, Path]) -> None:
         path.parent.mkdir(parents=True, exist_ok=True)
 
         data = self.model_dump(exclude_none=True, exclude_unset=True)
+        if "num_samples" in data:
+            data["num_samples"] = _sort_num_samples_keys(data["num_samples"])
         ordered_data = {key: data[key] for key in _KEY_ORDER if key in data}
 
         # Preserve any future fields.
diff --git a/src/plaid/storage/common/reader.py b/src/plaid/storage/common/reader.py
index dbca8407..ed7f5a75 100644
--- a/src/plaid/storage/common/reader.py
+++ b/src/plaid/storage/common/reader.py
@@ -54,7 +54,7 @@ def load_infos_from_disk(path: Union[str, Path]) -> Infos:
 
 def load_problem_definitions_from_disk(
     path: Union[str, Path],
-) -> dict[str, ProblemDefinition]:
+) -> dict[str, ProblemDefinition] | None:
     """Load ProblemDefinitions from a local dataset directory.
 
     This function reads all serialized ``ProblemDefinition`` files located in the
@@ -76,13 +76,9 @@ def load_problem_definitions_from_disk(
             Root dataset directory containing the ``problem_definitions/`` folder.
 
     Returns:
-        dict[str, ProblemDefinition]:
+        dict[str, ProblemDefinition] | None:
             Mapping from problem definition filename stems to loaded ``ProblemDefinition``
             objects.
-
-    Raises:
-        ValueError:
-            If the ``problem_definitions/`` directory does not exist.
     """
     pb_def_dir = Path(path).absolute()
     if pb_def_dir.name != "problem_definitions":
@@ -102,10 +98,6 @@ def load_problem_definitions_from_disk(
                     ) from exc
                 pb_defs[p.stem] = pb_def
         return pb_defs
-    else:
-        raise ValueError(
-            f"No problem definitions found on disk. path '{pb_def_dir}'"
-        )  # pragma: no cover
 
 
 def load_constants_from_disk(
diff --git a/src/plaid/storage/common/writer.py b/src/plaid/storage/common/writer.py
index fcc03232..85fec3c2 100644
--- a/src/plaid/storage/common/writer.py
+++ b/src/plaid/storage/common/writer.py
@@ -304,13 +304,16 @@ def push_local_problem_definitions_to_hub(
 
     api = HfApi()
 
-    api.upload_folder(
-        folder_path=path / Path("problem_definitions"),
-        repo_id=repo_id,
-        repo_type="dataset",
-        path_in_repo="problem_definitions",
-        commit_message="Upload problem_definitions",
-    )
+    try:
+        api.upload_folder(
+            folder_path=path / Path("problem_definitions"),
+            repo_id=repo_id,
+            repo_type="dataset",
+            path_in_repo="problem_definitions",
+            commit_message="Upload problem_definitions",
+        )
+    except ValueError:
+        logger.info(f"No problem definition in folder {path}")
 
 
 def push_local_metadata_to_hub(
diff --git a/tests/storage/test_storage.py b/tests/storage/test_storage.py
index e9727767..9bd5042b 100644
--- a/tests/storage/test_storage.py
+++ b/tests/storage/test_storage.py
@@ -289,8 +289,6 @@ def test_hf_datasets(
 
         loaded_pb_defs = load_problem_definitions_from_disk(test_dir)
         assert set(loaded_pb_defs) == {"pb_def"}
-        with pytest.raises(ValueError):
-            load_problem_definitions_from_disk("dummy")
 
         datasetdict, converterdict = init_from_disk(test_dir, splits=["train"])
         datasetdict, converterdict = init_from_disk(test_dir)
diff --git a/tests/test_info.py b/tests/test_info.py
index ae2d2e6d..b2ab6b65 100644
--- a/tests/test_info.py
+++ b/tests/test_info.py
@@ -170,6 +170,40 @@ def test_infos_save_and_load_roundtrip(tmp_path):
     assert reloaded.num_samples == {"train": 10}
 
 
+def test_infos_save_to_file_sorts_num_samples_keys(tmp_path):
+    model = Infos.model_validate(
+        _valid_infos(
+            num_samples={
+                "validation": 3,
+                "test_b": 2,
+                "train_b": 5,
+                "other": 1,
+                "test_a": 4,
+                "train_a": 6,
+                "train": 7,
+                "test": 8,
+            }
+        )
+    )
+
+    target = tmp_path / "infos.yaml"
+    model.save_to_file(target)
+    lines = target.read_text(encoding="utf-8").splitlines()
+
+    start = lines.index("num_samples:") + 1
+    end = lines.index("storage_backend: zarr")
+    assert lines[start:end] == [
+        "  train: 7",
+        "  train_a: 6",
+        "  train_b: 5",
+        "  test: 8",
+        "  test_a: 4",
+        "  test_b: 2",
+        "  other: 1",
+        "  validation: 3",
+    ]
+
+
 def test_infos_from_path_rejects_directory(tmp_path):
     Infos.model_validate(_valid_infos()).save_to_file(tmp_path / "infos.yaml")