diff --git a/inc/graphengine/inc/external/ge/ge_api_types.h b/inc/graphengine/inc/external/ge/ge_api_types.h
index 8efcef921a153c5d067d46dde6b88996c3699042..e2d720bef42f733a0cb7a472b1d84d105643004a 100644
--- a/inc/graphengine/inc/external/ge/ge_api_types.h
+++ b/inc/graphengine/inc/external/ge/ge_api_types.h
@@ -43,6 +43,13 @@ const char *const OPTION_EXEC_DEPLOY_MODE = "ge.exec.deployMode";
 const char *const OPTION_EXEC_RANK_TABLE_FILE = "ge.exec.rankTableFile";
 const char *const GE_AICPU_FLAG = "ge.aicpuFlag";
 const char *const OPTION_EXEC_EXTERN_PLUGIN_PATH = "ge.soLoadPath";
+
+const std::string OPTION_EXEC_CM_CHIEF_IP = "ge.cmChiefIp";
+const std::string OPTION_EXEC_CM_CHIEF_PORT = "ge.cmChiefPort";
+const std::string OPTION_EXEC_CM_CHIEF_DEVICE = "ge.cmChiefWorkerDevice";
+const std::string OPTION_EXEC_CM_WORKER_IP = "ge.cmWorkerIp";
+const std::string OPTION_EXEC_CM_WORKER_SIZE = "ge.cmWorkerSize";
+
 // Dump flag and para
 const char *const OPTION_EXEC_ENABLE_DUMP = "ge.exec.enableDump";
 const char *const OPTION_EXEC_DUMP_PATH = "ge.exec.dumpPath";
diff --git a/tf_adapter_2.x/npu_device/core/npu_wrapper.cpp b/tf_adapter_2.x/npu_device/core/npu_wrapper.cpp
index 7ffadcdd7b41d55992a34c5fab50dac8e6caabe0..7e15177529ffcabc8a843a456110df4c6f0f5c9f 100644
--- a/tf_adapter_2.x/npu_device/core/npu_wrapper.cpp
+++ b/tf_adapter_2.x/npu_device/core/npu_wrapper.cpp
@@ -117,7 +117,12 @@ const std::map<std::string, std::string> kConfigurableOptions = {
   {"graph_parallel_option_path", "ge.graphParallelOptionPath"},
   {"enable_graph_parallel", "ge.enableGraphParallel"},
   {"atomic_clean_policy", "ge.exec.atomicCleanPolicy"},
-  {"static_memory_policy", "ge.exec.staticMemoryPolicy"}};
+  {"static_memory_policy", "ge.exec.staticMemoryPolicy"},
+  {"_distribute.cm_chief_ip", ge::OPTION_EXEC_CM_CHIEF_IP},
+  {"_distribute.cm_chief_port", ge::OPTION_EXEC_CM_CHIEF_PORT},
+  {"_distribute.cm_chief_worker_device", ge::OPTION_EXEC_CM_CHIEF_DEVICE},
+  {"_distribute.cm_worker_ip", ge::OPTION_EXEC_CM_WORKER_IP},
+  {"_distribute.cm_worker_size", ge::OPTION_EXEC_CM_WORKER_SIZE}};
 }  // namespace
 
 #undef PYBIND11_CHECK_PYTHON_VERSION
@@ -147,6 +152,16 @@ void ParseGlobalOptions(int device_index, const std::map<std::string, std::strin
     global_options["ge.exec.hccl_tailing_optimize"] = kTrue;
   }
 
+  if (global_options.find(ge::OPTION_EXEC_CM_CHIEF_IP) != global_options.end() &&
+      global_options.find(ge::OPTION_EXEC_CM_CHIEF_PORT) != global_options.end() &&
+      global_options.find(ge::OPTION_EXEC_CM_CHIEF_DEVICE) != global_options.end()) {
+    const static std::string kTrue = "1";
+    global_options[ge::OPTION_EXEC_DEPLOY_MODE] = "0";
+    global_options[ge::OPTION_EXEC_IS_USEHCOM] = kTrue;
+    global_options[ge::OPTION_EXEC_HCCL_FLAG] = kTrue;
+    global_options["ge.exec.hccl_tailing_optimize"] = kTrue;
+  }
+
   if (global_options.find("ge.jobType") == global_options.end()) {
     if (!kAoeMode.empty()) {
       global_options["ge.jobType"] = kAoeMode;
diff --git a/tf_adapter_2.x/python/npu_device/distribute/npu_callbacks.py b/tf_adapter_2.x/python/npu_device/distribute/npu_callbacks.py
index bfafab2dc2ac92c58e4bcd16e71b9475a7010fda..3b359265c484bd75a12a0259c0f79fe77a7a2540 100644
--- a/tf_adapter_2.x/python/npu_device/distribute/npu_callbacks.py
+++ b/tf_adapter_2.x/python/npu_device/distribute/npu_callbacks.py
@@ -52,9 +52,9 @@ def broadcast_keras_model(model, root_rank=0):
 
 
 def get_ranksize():
-    if os.getenv("CM_WORK_SIZE") is not None and os.getenv("RANK_SIZE") is not None:
+    if os.getenv("CM_WORKER_SIZE") is not None and os.getenv("RANK_SIZE") is not None:
         raise ValueError("RANK_SIZE and CM_WORK_SIZE cannot be configured at the same time")
-    rank_size = os.getenv('RANK_SIZE') if os.getenv("RANK_SIZE") is not None else os.getenv('CM_WORK_SIZE', '1')
+    rank_size = os.getenv('RANK_SIZE') if os.getenv("RANK_SIZE") is not None else os.getenv('CM_WORKER_SIZE', '1')
     return rank_size
 
 
diff --git a/tf_adapter_2.x/python/npu_device/npu_device.py b/tf_adapter_2.x/python/npu_device/npu_device.py
index d7bca74dbbc4b1ac533a7070d9feb34fd220649b..053c40f7f4a6c85531ba022090cd9b98d24b3ba4 100644
--- a/tf_adapter_2.x/python/npu_device/npu_device.py
+++ b/tf_adapter_2.x/python/npu_device/npu_device.py
@@ -131,6 +131,38 @@ def enable_v1():
     tf.load_op_library(os.path.join(os.path.dirname(__file__), "compat", "v1", "_tf_adapter.so"))
 
 
+def get_ranksize():
+    if os.getenv("CM_WORKER_SIZE") is not None and os.getenv("RANK_SIZE") is not None:
+        raise ValueError("RANK_SIZE and CM_WORK_SIZE cannot be configured at the same time")
+    rank_size = os.getenv('RANK_SIZE') if os.getenv("RANK_SIZE") is not None else os.getenv('CM_WORKER_SIZE', '1')
+    return rank_size
+
+
+def set_cm_chief_worksize_env(global_kw_options=None, env_cm_chief_ip=None, workers_num=None):
+    env_cm_chief_port = os.getenv("CM_CHIEF_PORT")
+    env_cm_chief_device = os.getenv("CM_CHIEF_DEVICE")
+    env_cm_worker_ip = os.getenv("CM_WORKER_IP")
+    if not env_cm_chief_port:
+        raise RuntimeError('You must specify cm chief port by set env CM_CHIEF_PORT in distribution mode')
+    if not env_cm_chief_device:
+        raise RuntimeError('You must specify cm chief device by set env CM_CHIEF_DEVICE in distribution mode')
+    if not env_cm_chief_ip:
+        raise RuntimeError('You must specify cm chief ip by set env CM_CHIEF_IP in distribution mode')
+    global_kw_options['_distribute.cm_chief_ip'] = env_cm_chief_ip
+    global_kw_options['_distribute.cm_chief_port'] = env_cm_chief_port
+    global_kw_options['_distribute.cm_chief_worker_device'] = env_cm_chief_device
+    global_kw_options['_distribute.cm_worker_ip'] = env_cm_worker_ip
+    global_kw_options['_distribute.cm_worker_size'] = str(workers_num)
+
+
+def set_rank_table_file_env(global_kw_options=None, env_rank_table=None):
+    env_worker_id = os.getenv('RANK_ID')
+    if not env_worker_id:
+        raise RuntimeError('You must specify rank id by set env RANK_ID in distribution mode')
+    global_kw_options['_distribute.rank_table'] = env_rank_table
+    global_kw_options['_distribute.rank_id'] = env_worker_id
+
+
 def open(device_id=None):
     """Initiate and return a NPU device handle"""
     if device_id is None:
@@ -152,18 +184,18 @@ def open(device_id=None):
                                ''.format(device_id, list(_npu_device_instances.keys())))
 
         global_kw_options = global_options().as_dict()
-        workers_num = int(os.getenv('RANK_SIZE', '1'))
+        workers_num = int(get_ranksize())
         if workers_num > 1:
+            env_cm_chief_ip = os.getenv("CM_CHIEF_IP")
             env_rank_table = os.getenv("RANK_TABLE_FILE")
-            env_worker_id = os.getenv('RANK_ID')
-            if not env_rank_table:
-                raise RuntimeError('You must specify a rank table file by set env RANK_TABLE_FILE in distribution mode')
-
-            if not env_worker_id:
-                raise RuntimeError('You must specify rank id by set env RANK_ID in distribution mode')
-
-            global_kw_options['_distribute.rank_table'] = env_rank_table
-            global_kw_options['_distribute.rank_id'] = env_worker_id
+            if env_cm_chief_ip is not None and env_rank_table is not None:
+                raise RuntimeError('CM_CHIEF_IP and RANK_TABLE_FILE cannot be configured at the same time.')
+            elif env_cm_chief_ip is not None:
+                set_cm_chief_worksize_env(global_kw_options, env_cm_chief_ip, workers_num)
+            elif env_rank_table is not None:
+                set_rank_table_file_env(global_kw_options, env_rank_table)
+            else:
+                raise RuntimeError('CM_CHIEF_IP and RANK_TABLE_FILE are all not be configured.')
 
         device_options = {}
         error_message = _npu_device_backends.Open(context.context()._handle, NPU, device_id, global_kw_options,
diff --git a/tf_adapter_2.x/tests/st/adapter2_st.py b/tf_adapter_2.x/tests/st/adapter2_st.py
index 1b28a15cba38c9c6e7bc8a7c740425f735ce74e6..03663ac704bf1ef0195b8e05a01ea9c7eb66b628 100644
--- a/tf_adapter_2.x/tests/st/adapter2_st.py
+++ b/tf_adapter_2.x/tests/st/adapter2_st.py
@@ -35,9 +35,11 @@ npu_device.global_options().experimental.multi_branches_config.dynamic_node_type
 npu_device.global_options().experimental.multi_branches_config.dynamic_dims = "1;2"
 npu_device.global_options().aoe_config.work_path = "./"
 npu_device.global_options().graph_run_mode = 0
-os.environ['RANK_TABLE_FILE'] = "rankTable"
-os.environ['RANK_SIZE'] = "2"
-os.environ['RANK_ID'] = "1"
+os.environ['CM_CHIEF_IP'] = "1"
+os.environ['CM_CHIEF_PORT'] = "3"
+os.environ['CM_CHIEF_DEVICE'] = "4"
+os.environ['CM_WORKER_SIZE'] = "2"
+os.environ['CM_WORKER_IP'] = "123"
 npu = npu_device.open().as_default()
 npu.workers_num = 2  # mock run in 2P env
 
diff --git a/tf_adapter_2.x/tests/stub/include/stub/defines.h b/tf_adapter_2.x/tests/stub/include/stub/defines.h
index 5be2c1243507db9bf8e99b17af920d5bac39060d..653de3c59dc2b81e91fda1a7704b091bca4d4ae4 100644
--- a/tf_adapter_2.x/tests/stub/include/stub/defines.h
+++ b/tf_adapter_2.x/tests/stub/include/stub/defines.h
@@ -25,6 +25,13 @@ const char *const OPTION_EXEC_DEPLOY_MODE = "ge.exec.deployMode";
 const char *const OPTION_EXEC_RANK_TABLE_FILE = "ge.exec.rankTableFile";
 const char *const GE_AICPU_FLAG = "ge.aicpuFlag";
 const char *const OPTION_EXEC_EXTERN_PLUGIN_PATH = "ge.soLoadPath";
+
+const std::string OPTION_EXEC_CM_CHIEF_IP = "ge.cmChiefIp";
+const std::string OPTION_EXEC_CM_CHIEF_PORT = "ge.cmChiefPort";
+const std::string OPTION_EXEC_CM_CHIEF_DEVICE = "ge.cmChiefWorkerDevice";
+const std::string OPTION_EXEC_CM_WORKER_IP = "ge.cmWorkerIp";
+const std::string OPTION_EXEC_CM_WORKER_SIZE = "ge.cmWorkerSize";
+
 // Dump flag and para
 const char *const OPTION_EXEC_ENABLE_DUMP = "ge.exec.enableDump";
 const char *const OPTION_EXEC_DUMP_PATH = "ge.exec.dumpPath";