diff --git a/mineru/backend/vlm/utils.py b/mineru/backend/vlm/utils.py
index a5aac6e3..162adc67 100644
--- a/mineru/backend/vlm/utils.py
+++ b/mineru/backend/vlm/utils.py
@@ -100,4 +100,19 @@ def set_default_batch_size() -> int:
     except Exception as e:
         logger.warning(f'Error determining VRAM: {e}, using default batch_ratio: 1')
         batch_size = 1
-    return batch_size
\ No newline at end of file
+    return batch_size
+
+
+def set_compilation_config() -> dict:
+    device = get_device()
+    compilation_config = {}
+    if str(device).startswith('musa'):
+        try:
+            import torch
+            if torch.musa.is_available():
+                compilation_config = {
+                    "simple_cuda_graph": true
+                }
+        except Exception as e:
+            pass
+    return compilation_config
\ No newline at end of file
diff --git a/mineru/backend/vlm/vlm_analyze.py b/mineru/backend/vlm/vlm_analyze.py
index ca7373d4..2a9e6eb5 100644
--- a/mineru/backend/vlm/vlm_analyze.py
+++ b/mineru/backend/vlm/vlm_analyze.py
@@ -99,6 +99,20 @@ class ModelSingleton:
                         import vllm
                     except ImportError:
                         raise ImportError("Please install vllm to use the vllm-engine backend.")
+
+                    if "compilation_config" in kwargs:
+                        if isinstance(kwargs["compilation_config"], str):
+                            try:
+                                kwargs["compilation_config"] = json.loads(kwargs["compilation_config"])
+                            except json.JSONDecodeError:
+                                logger.warning(
+                                    f"Failed to parse compilation_config as JSON: {kwargs['compilation_config']}")
+                                del kwargs["compilation_config"]
+                    else:
+                        compilation_config = set_compilation_config()
+                        if compilation_config:
+                            kwargs["compilation_config"] = compilation_config
+
                     if "gpu_memory_utilization" not in kwargs:
                         kwargs["gpu_memory_utilization"] = set_default_gpu_memory_utilization()
                     if "model" not in kwargs:
@@ -114,6 +128,20 @@ class ModelSingleton:
                         from vllm.v1.engine.async_llm import AsyncLLM
                     except ImportError:
                         raise ImportError("Please install vllm to use the vllm-async-engine backend.")
+
+                    if "compilation_config" in kwargs:
+                        if isinstance(kwargs["compilation_config"], str):
+                            try:
+                                kwargs["compilation_config"] = json.loads(kwargs["compilation_config"])
+                            except json.JSONDecodeError:
+                                logger.warning(
+                                    f"Failed to parse compilation_config as JSON: {kwargs['compilation_config']}")
+                                del kwargs["compilation_config"]
+                    else:
+                        compilation_config = set_compilation_config()
+                        if compilation_config:
+                            kwargs["compilation_config"] = compilation_config
+
                     if "gpu_memory_utilization" not in kwargs:
                         kwargs["gpu_memory_utilization"] = set_default_gpu_memory_utilization()
                     if "model" not in kwargs: