[ModelOpt FP4] Support online MoE weight quantization (#33115)

Co-authored-by: Brayden Zhong <b8zhong@uwaterloo.ca>

test/registered/unit/server_args/test_server_args.py

@@ -1,3 +1,4 @@
+import dataclasses
 import importlib
 import json
 import os
@@ -81,6 +82,17 @@ class TestPrepareServerArgs(CustomTestCase):
                 return_hidden_states_mode="lst",
             )
 
+    def test_draft_quantization_explicitness_survives_asdict_round_trip(self):
+        inherited = ServerArgs(model_path="dummy", quantization="modelopt_fp4")
+        inherited._handle_missing_default_values()
+        self.assertEqual(inherited.speculative_draft_model_quantization, "modelopt_fp4")
+        self.assertFalse(inherited._speculative_draft_quantization_explicitly_set)
+
+        reconstructed = ServerArgs(**dataclasses.asdict(inherited))
+        reconstructed._handle_missing_default_values()
+
+        self.assertFalse(reconstructed._speculative_draft_quantization_explicitly_set)
+
     def test_config_nested_dict_args_are_json(self):
         with tempfile.NamedTemporaryFile(mode="w", suffix=".yaml", delete=False) as f:
             f.write("mm-process-config:\n  image:\n    resize: 128\n")