diff --git a/vllm_mindspore/__init__.py b/vllm_mindspore/__init__.py
index 5892937a8957829c8a47ae1df13ada581137f7e6..8bd06a48501f9c6a2ccfd7959eae34d6eb57dec1 100644
--- a/vllm_mindspore/__init__.py
+++ b/vllm_mindspore/__init__.py
@@ -59,6 +59,19 @@ import vllm.v1.engine.core
 from vllm_mindspore.v1.engine.core import shutdown
 vllm.v1.engine.core.DPEngineCoreProc.shutdown = shutdown
 
+import vllm.lora.utils
+from vllm_mindspore.lora.utils import get_supported_lora_modules_ms
+vllm.lora.utils.get_supported_lora_modules = get_supported_lora_modules_ms
+
+from vllm_mindspore.lora.models import LoRAModel_ms, LoRAModelManager_ms
+from vllm_mindspore.lora.worker_manager import _load_adapter_ms    
+import LoRAModelManager
+from vllm.lora.worker_manager import WorkerLoRAManager
+LoRAModelManager._create_lora_modules = LoRAModelManager_ms._create_lora_modules
+LoRAModelManager._set_adapter_mapping = LoRAModelManager_ms._set_adapter_mapping
+WorkerLoRAManager._load_adapter = _load_adapter_ms
+vllm.lora.models.LoRAModel = LoRAModel_ms
+
 from vllm_mindspore.utils import (
     direct_register_custom_op,
     make_tensor_with_pad,
diff --git a/vllm_mindspore/lora/__init__.py b/vllm_mindspore/lora/__init__.py
new file mode 100644
index 0000000000000000000000000000000000000000..e69de29bb2d1d6434b8b29ae775ad8c2e48c5391
diff --git a/vllm_mindspore/lora/models.py b/vllm_mindspore/lora/models.py
new file mode 100644
index 0000000000000000000000000000000000000000..2b2073ca1f188c625215f36b396d68cdc47c6847
--- /dev/null
+++ b/vllm_mindspore/lora/models.py
@@ -0,0 +1,11 @@
+from vllm.lora.models import LoRAModelManager, LoRAModel
+from vllm_mindspore.utils import is_mindformers_model_backend
+from mindformers.parallel_core.inference.lora.models import LoRAModel as MFLoRAModel
+from mindformers.parallel_core.inference.lora.models import vLLMLoRAModelManager
+
+if is_mindformers_model_backend():
+    LoRAModel_ms = MFLoRAModel
+    LoRAModelManager_ms = vLLMLoRAModelManager
+else:
+    LoRAModel_ms = LoRAModel
+    LoRAModelManager_ms = LoRAModelManager
\ No newline at end of file
diff --git a/vllm_mindspore/lora/utils.py b/vllm_mindspore/lora/utils.py
new file mode 100644
index 0000000000000000000000000000000000000000..d8c45057e5e11af082a23871aa2ccc605a8ed93d
--- /dev/null
+++ b/vllm_mindspore/lora/utils.py
@@ -0,0 +1,8 @@
+from vllm.lora.utils import get_supported_lora_modules
+from vllm_mindspore.utils import is_mindformers_model_backend
+from mindformers.parallel_core.inference.lora.lora_utils import get_mf_supported_lora_modules
+
+if is_mindformers_model_backend():
+    get_supported_lora_modules_ms = get_mf_supported_lora_modules
+else:
+    get_supported_lora_modules_ms = get_supported_lora_modules
\ No newline at end of file
diff --git a/vllm_mindspore/lora/worker_manager.py b/vllm_mindspore/lora/worker_manager.py
new file mode 100644
index 0000000000000000000000000000000000000000..bc86937b9b0b9932309505d4ab75657a5fc4d78e
--- /dev/null
+++ b/vllm_mindspore/lora/worker_manager.py
@@ -0,0 +1,31 @@
+from typing import List
+from vllm.lora.request import LoRARequest
+from vllm_mindspore.utils import is_mindformers_model_backend
+from mindformers.parallel_core.inference.lora.models import load_lora_ckpt
+
+def _load_adapter_ms(self, lora_request: LoRARequest):
+    supported_lora_modules = (
+        self._adapter_manager.supported_lora_modules)
+    packed_modules_mapping = (
+        self._adapter_manager.packed_modules_mapping)
+    expected_lora_modules: List[str] = []
+    for module in supported_lora_modules:
+        if module in packed_modules_mapping:
+            expected_lora_modules.extend(
+                packed_modules_mapping[module])
+    else:
+        expected_lora_modules.append(module)
+
+    expected_lora_modules = list(set(expected_lora_modules))
+    lora = load_lora_ckpt(lora_request, 
+                        self.lora_config, 
+                        self.vocab_size, 
+                        expected_lora_modules, 
+                        self.max_position_embeddings, 
+                        self.embedding_modules, 
+                        self.embedding_padding_modules)
+    return lora
+
+if not is_mindformers_model_backend():
+    from vllm.lora.worker_manager import WorkerLoRAManager
+    _load_adapter_ms = WorkerLoRAManager._load_adapter
\ No newline at end of file