diff --git a/vllm/lora/lora_model.py b/vllm/lora/lora_model.py
index f5e36697e..bc88c71ea 100644
--- a/vllm/lora/lora_model.py
+++ b/vllm/lora/lora_model.py
@@ -12,7 +12,6 @@ from vllm.lora.peft_helper import PEFTHelper
 from vllm.lora.utils import (
     get_lora_id,
     is_base_embeddding_weights,
-    is_regex_target_modules,
     parse_fine_tuned_lora_name,
 )
 from vllm.model_executor.model_loader.tensorizer import TensorizerConfig
@@ -201,37 +200,13 @@ class LoRAModel:
                 for module in f.keys():  # noqa
                     tensors[module] = f.get_tensor(module)
         elif os.path.isfile(lora_bin_file_path) or os.path.isfile(lora_pt_file_path):
-            # When a bin/pt file is provided, we rely on config to find
-            # unexpected modules.
-            unexpected_modules = []
-            target_modules = peft_helper.target_modules
-            if not isinstance(target_modules, list):
-                target_modules = [target_modules]
-            for module in target_modules:
-                # Compatible with more modules,
-                # such as:layers.11.self_attn.k_proj
-                part_name = module.split(".")[-1]
-                if part_name not in expected_lora_modules:
-                    unexpected_modules.append(module)
-            # loaded lora's target modules must be a subset of
-            # expected_lora_modules. It is not reliable. See
-            # https://github.com/vllm-project/vllm/pull/5909. But there's no
-            # other better mechanism.
-            if unexpected_modules and not is_regex_target_modules(
-                peft_helper.target_modules, expected_lora_modules
-            ):
-                raise ValueError(
-                    f"While loading {lora_dir}, expected"
-                    f" target modules in {expected_lora_modules}"
-                    f" but received {unexpected_modules}."
-                    f" Please verify that the loaded LoRA module is correct"
-                )
             lora_file_path = (
                 lora_bin_file_path
                 if os.path.isfile(lora_bin_file_path)
                 else lora_pt_file_path
             )
             tensors = torch.load(lora_file_path, map_location=device, weights_only=True)
+            check_unexpected_modules(tensors)
         else:
             raise ValueError(f"{lora_dir} doesn't contain tensors")
 
diff --git a/vllm/lora/utils.py b/vllm/lora/utils.py
index 4d264c068..75aeccd00 100644
--- a/vllm/lora/utils.py
+++ b/vllm/lora/utils.py
@@ -5,7 +5,6 @@ import os
 from typing import TYPE_CHECKING, Optional
 
 import huggingface_hub
-import regex as re
 from huggingface_hub.utils import (
     EntryNotFoundError,
     HfHubHTTPError,
@@ -186,39 +185,6 @@ def is_base_embeddding_weights(name: str) -> bool:
     return name.endswith(embedding_suffixes)
 
 
-def is_regex_target_modules(
-    load_modules: str | list[str], expected_lora_modules: set[str]
-) -> bool:
-    """
-    PEFT supports passing `target_modules` in the form of regular expressions,
-    such as `model.*(q_proj|k_proj|v_proj)$`. This function is mainly used to
-    determine whether the suffix in the regular expression is present in the
-    `expected_lora_modules`.
-    """
-
-    def is_valid_regex(pattern):
-        try:
-            re.compile(pattern)
-            return True
-        except re.error:
-            return False
-
-    def is_subset(sub_list, full_set):
-        return set(sub_list).issubset(full_set)
-
-    # Similar to PEFT's processing logic, regex-related operations are only
-    #  executed when the load_modules is a `str`.
-    if not isinstance(load_modules, str):
-        return False
-
-    if is_valid_regex(load_modules):
-        match = re.search(r"\((.*?)\)\$?$", load_modules)
-        if match:
-            suffix = match.group(1).split("|")
-            return is_subset(suffix, expected_lora_modules)
-    return False
-
-
 def get_supported_lora_modules(model: nn.Module) -> list[str]:
     """
     In vLLM, all linear layers support LoRA.