diff --git a/python/sglang/srt/managers/router/model_runner.py b/python/sglang/srt/managers/router/model_runner.py
index 6f52db83d..afbe03abe 100644
--- a/python/sglang/srt/managers/router/model_runner.py
+++ b/python/sglang/srt/managers/router/model_runner.py
@@ -18,6 +18,9 @@ from vllm.model_executor.layers.quantization.marlin import MarlinConfig
 from vllm.model_executor.model_loader import _set_default_torch_dtype
 from vllm.model_executor.parallel_utils.parallel_state import initialize_model_parallel
 
+import importlib
+import pkgutil
+
 import sglang
 
 QUANTIONCONFIG_MAPPING = {"awq": AWQConfig, "gptq": GPTQConfig, "marlin": MarlinConfig}
@@ -32,10 +35,11 @@ global_server_args_dict: dict = None
 @lru_cache()
 def import_model_classes():
     model_arch_name_to_cls = {}
-    for f in importlib.resources.files("sglang.srt.models").iterdir():
-        if f.name.endswith(".py"):
-            module_name = Path(f.name).with_suffix('')
-            module = importlib.import_module(f"sglang.srt.models.{module_name}")
+    package_name = "sglang.srt.models"
+    package = importlib.import_module(package_name)
+    for finder, name, ispkg in pkgutil.iter_modules(package.__path__, package_name + '.'):
+        if not ispkg:
+            module = importlib.import_module(name)
             if hasattr(module, "EntryClass"):
                 model_arch_name_to_cls[module.EntryClass.__name__] = module.EntryClass
     return model_arch_name_to_cls