[quantization][config] enable override existing quant_config (#28510)

Signed-off-by: Hank <hcc.mayday@gmail.com>
Co-authored-by: Michael Goin <mgoin64@gmail.com>

tests/quantization/test_register_quantization_config.py

@@ -7,6 +7,7 @@ See https://github.com/vllm-project/vllm/issues/11926 for more details.
 Run `pytest tests/quantization/test_register_quantization_config.py`.
 """
 
+import logging
 from typing import Any
 
 import pytest
@@ -100,17 +101,22 @@ class CustomQuantConfig(QuantizationConfig):
         return None
 
 
-def test_register_quantization_config():
+def test_register_quantization_config(caplog_vllm):
     """Test register custom quantization config."""
 
     # The quantization method `custom_quant` should be registered.
     assert get_quantization_config("custom_quant") == CustomQuantConfig
 
     # The quantization method `custom_quant` is already exists,
-    # should raise an error.
+    # should raise a warning when re-registering it.
-    with pytest.raises(ValueError):
+    with caplog_vllm.at_level(logging.WARNING):
         register_quantization_config("custom_quant")(CustomQuantConfig)
 
+    assert any(
+        "The quantization method 'custom_quant' already exists" in message
+        for message in caplog_vllm.messages
+    ), "Expected a warning when re-registering custom_quant"
+
 
 @pytest.mark.parametrize(
     argnames="model",

vllm/model_executor/layers/quantization/__init__.py

@@ -3,8 +3,11 @@
 
 from typing import Literal, get_args
 
+from vllm.logger import init_logger
 from vllm.model_executor.layers.quantization.base_config import QuantizationConfig
 
+logger = init_logger(__name__)
+
 QuantizationMethods = Literal[
     "awq",
     "deepspeedfp",
@@ -70,15 +73,20 @@ def register_quantization_config(quantization: str):
 
     def _wrapper(quant_config_cls):
         if quantization in QUANTIZATION_METHODS:
-            raise ValueError(
+            logger.warning(
-                f"The quantization method `{quantization}` is already exists."
+                "The quantization method '%s' already exists and will be "
+                "overwritten by the quantization config %s.",
+                quantization,
+                quant_config_cls,
             )
+        else:
+            QUANTIZATION_METHODS.append(quantization)
+
         if not issubclass(quant_config_cls, QuantizationConfig):
             raise ValueError(
                 "The quantization config must be a subclass of `QuantizationConfig`."
             )
         _CUSTOMIZED_METHOD_TO_QUANT_CONFIG[quantization] = quant_config_cls
-        QUANTIZATION_METHODS.append(quantization)
         return quant_config_cls
 
     return _wrapper