Test warning raised on unmatched target

Author: fynnsu

tests/test_quantization/lifecycle/test_apply.py

@@ -258,13 +258,15 @@ def get_sample_tinyllama_quant_config(status: str = "frozen"):
 
 @requires_accelerate()
 @pytest.mark.parametrize(
-    "ignore",
+    "target,should_raise_warning",
     [
-        ("lm_head", "re:.*gate"),
-        ("lm_head", "re:.*foobarbaz"),
+        [("Linear",), False],
+        [("Linear", "re:.*foobarbaz"), True],
     ],
 )
-def test_apply_quantization_status(ignore):
+def test_apply_quantization_status(caplog, target, should_raise_warning):
+    import logging
+
     # load a dense, unquantized tiny llama model
     model = get_tinyllama_model()
     quantization_config_dict = {
@@ -279,13 +281,19 @@ def test_apply_quantization_status(ignore):
                     "symmetric": False,
                     "strategy": "tensor",
                 },
-                "targets": ["Linear"],
+                "targets": target,
             }
         },
+        "ignore": ["lm_head", "re:.*gate"],
     }
-    quantization_config_dict["ignore"] = ignore
 
     config = QuantizationConfig(**quantization_config_dict)
     config.quantization_status = QuantizationStatus.CALIBRATION
 
-    apply_quantization_config(model, config)
+    # mismatch in the ignore key of quantization_config_dict
+    with caplog.at_level(logging.WARNING):
+        apply_quantization_config(model, config)
+        if should_raise_warning:
+            assert len(caplog.text) > 0
+        else:
+            assert len(caplog.text) == 0