Ignore unrelated files during model version discovery

great_ai/large_file/large_file/large_file_base.py

@@ -273,21 +273,15 @@ class LargeFileBase(ABC):
     def _find_instances_from_cache(self) -> List[DataInstance]:
         logger.info(f"Fetching cached versions of {self._name}")
 
-        candidates = [
+        prefix = f"{self._name}{CACHE_NAME_VERSION_SEPARATOR}"
+        return [
             DataInstance(
-                name=CACHE_NAME_VERSION_SEPARATOR.join(
+                name=self._name, version=int(f.name[len(prefix) :]), remote_path=f
-                    f.name.split(CACHE_NAME_VERSION_SEPARATOR)[:-1]
-                ),
-                version=int(f.name.split(CACHE_NAME_VERSION_SEPARATOR)[-1]),
-                remote_path=f,
-            )
-            for f in self.cache_path.glob(
-                f"{self._name}{CACHE_NAME_VERSION_SEPARATOR}*"
             )
+            for f in self.cache_path.glob(f"{prefix}*")
+            if f.name[len(prefix) :].isdigit()
         ]
 
-        return [c for c in candidates if c.name == self._name]
-
     def _check_mode_and_set_version(self) -> None:
         if "+" in self._mode:
             raise ValueError(

tests/large_file/test_storage_safety.py

@@ -20,3 +20,10 @@ def test_configured_cache_directory_is_created(
     with LargeFileLocal("model", "w") as file:
         file.write("weights")
     assert LargeFileLocal("model").get().read_text() == "weights"
+
+
+def test_unrelated_cache_entries_are_ignored() -> None:
+    LargeFileLocal.cache_path.mkdir(parents=True)
+    (LargeFileLocal.cache_path / "model-unrelated").write_text("unrelated")
+    (LargeFileLocal.cache_path / "model-0").write_text("weights")
+    assert LargeFileLocal("model").get().read_text() == "weights"