fix run error

vllm/forward_context.py

@@ -210,4 +210,13 @@ def set_profilling(profiling):
 
 def get_profilling() -> bool:
     global _profiling
-    return _profiling
\ No newline at end of file
+    return _profiling
+
+@contextmanager
+def set_warming_up(warming_up):
+    global _warming_up
+    _warming_up = warming_up
+
+def get_warming_up() -> bool:
+    global _warming_up
+    return _warming_up
\ No newline at end of file

vllm/model_executor/layers/linear.py

@@ -1578,6 +1578,11 @@ class RowParallelLinear(LinearBase):
         # Divide the weight matrix along the first dimension.
         self.tp_rank = get_tensor_model_parallel_rank()
         self.tp_size = get_tensor_model_parallel_world_size()
+        
+        self.enable_dp_attn_moe = enable_dp_attn_moe
+        if enable_dp_attn_moe:
+            self.tp_rank = get_moe_tp_rank()
+            self.tp_size = get_moe_tp_size()
 
         if expect_tp_size is not None:
             self.tp_rank = 0