[quantization] Fix internal test

test/quantization/algorithm/test_gptq.py

@@ -391,7 +391,7 @@ def test_model(self):
         )
 
         # Load data
-        dataset = load_dataset("wikiText", "wikitext-2-raw-v1", split="train")
+        dataset = load_dataset("wikitext", "wikitext-2-raw-v1", split="train")
         sample_input = tokenizer(dataset[0]["text"], return_tensors="pt").input_ids
 
         # base

test/quantization/algorithm/test_smooth_quant.py

@@ -49,7 +49,7 @@ def test_value(self):
         )
 
         # Load data
-        dataset = load_dataset("wikiText", "wikitext-2-raw-v1", split="train")
+        dataset = load_dataset("wikitext", "wikitext-2-raw-v1", split="train")
         sample_input = tokenizer(dataset[0]["text"], return_tensors="pt").input_ids
 
         # base

test/quantization/recipes/optional_dependency_stubs.py

@@ -21,29 +21,69 @@
 make the tests heavier than necessary.
 """
 
+import importlib
 import sys
 import types
 
+_STUB_MARKER = "__tico_optional_dependency_stub__"
+
 
 def install_optional_dependency_stubs() -> None:
     """Install lightweight stubs for optional recipe dependencies."""
     _install_datasets_stub()
     _install_lm_eval_stub()
 
 
+def _has_attrs(module: types.ModuleType, names: tuple[str, ...]) -> bool:
+    """Return True if a module has all required attributes."""
+    return all(hasattr(module, name) for name in names)
+
+
+def _is_our_stub(module: types.ModuleType | None) -> bool:
+    """Return True if a module was installed by this stub helper."""
+    return bool(module is not None and getattr(module, _STUB_MARKER, False))
+
+
+def _try_import_optional_module(module_name: str) -> types.ModuleType | None:
+    """
+    Import an optional module if it is available.
+
+    Missing optional modules are tolerated. Import failures caused by missing
+    transitive dependencies are re-raised so broken real installations are not
+    silently hidden by test stubs.
+    """
+    try:
+        return importlib.import_module(module_name)
+    except ModuleNotFoundError as exc:
+        top_level_name = module_name.partition(".")[0]
+        if exc.name in {module_name, top_level_name}:
+            return None
+        raise
+
+
 def _install_datasets_stub() -> None:
     """Install a minimal datasets module when the real package is unavailable."""
-    if "datasets" in sys.modules and all(
-        hasattr(sys.modules["datasets"], name)
-        for name in ("Dataset", "IterableDataset", "load_dataset")
+    required_attrs = ("Dataset", "IterableDataset", "load_dataset")
+    existing_module = sys.modules.get("datasets")
+
+    if (
+        existing_module is not None
+        and not _is_our_stub(existing_module)
+        and _has_attrs(existing_module, required_attrs)
     ):
         return
 
+    real_module = _try_import_optional_module("datasets")
+    if real_module is not None and _has_attrs(real_module, required_attrs):
+        return
+
     module = sys.modules.get("datasets")
-    if module is None:
+    if module is None or not _is_our_stub(module):
         module = types.ModuleType("datasets")
         sys.modules["datasets"] = module
 
+    setattr(module, _STUB_MARKER, True)
+
     class Dataset:
         """Minimal datasets.Dataset stub for import-time compatibility."""
 
@@ -74,15 +114,31 @@ def load_dataset(*args, **kwargs):
 
 def _install_lm_eval_stub() -> None:
     """Install minimal lm_eval modules when the real package is unavailable."""
-    if "lm_eval" in sys.modules and hasattr(sys.modules["lm_eval"], "evaluator"):
+    existing_module = sys.modules.get("lm_eval")
+
+    if (
+        existing_module is not None
+        and not _is_our_stub(existing_module)
+        and hasattr(existing_module, "evaluator")
+    ):
         return
 
+    real_module = _try_import_optional_module("lm_eval")
+    if real_module is not None and not _is_our_stub(real_module):
+        real_evaluator_module = _try_import_optional_module("lm_eval.evaluator")
+        if real_evaluator_module is not None:
+            setattr(real_module, "evaluator", real_evaluator_module)
+            return
+
     lm_eval_module = sys.modules.get("lm_eval")
-    if lm_eval_module is None:
+    if lm_eval_module is None or not _is_our_stub(lm_eval_module):
         lm_eval_module = types.ModuleType("lm_eval")
         sys.modules["lm_eval"] = lm_eval_module
 
+    setattr(lm_eval_module, _STUB_MARKER, True)
+
     evaluator_module = types.ModuleType("lm_eval.evaluator")
+    setattr(evaluator_module, _STUB_MARKER, True)
 
     def simple_evaluate(*args, **kwargs):
         """Fail clearly if a test accidentally runs real lm-eval."""
@@ -96,6 +152,7 @@ def simple_evaluate(*args, **kwargs):
     setattr(lm_eval_module, "evaluator", evaluator_module)
 
     utils_module = types.ModuleType("lm_eval.utils")
+    setattr(utils_module, _STUB_MARKER, True)
 
     def make_table(results):
         """Return a stable string representation for patched evaluation results."""
@@ -106,7 +163,10 @@ def make_table(results):
     setattr(lm_eval_module, "utils", utils_module)
 
     models_module = types.ModuleType("lm_eval.models")
+    setattr(models_module, _STUB_MARKER, True)
+
     huggingface_module = types.ModuleType("lm_eval.models.huggingface")
+    setattr(huggingface_module, _STUB_MARKER, True)
 
     class HFLM:
         """Minimal HFLM stub for import-time compatibility."""