Upd

Author: wenscarl

flashinfer/comm/mnnvl.py

@@ -547,7 +547,6 @@ def supports_mnnvl() -> bool:
 
 class McastDeviceMemory:
     """Python port of McastDeviceMemory from TensorRT-LLM"""
-    # config: Optional[MnnvlConfig] = None
     def __init__(
         self,
         buf_size: int,
@@ -650,7 +649,6 @@ def __init__(
         self.signal_pads_dev = alloc_and_copy_to_cuda(self.signal_pads)
         self.uc_ptrs_dev = alloc_and_copy_to_cuda(self.uc_ptrs)
 
-
     def __del__(self):
         """Destructor - cleanup allocated memory"""
 
@@ -769,14 +767,6 @@ def _alloc_mn_mcast_mem(self, buf_size: int, comm: Any=MpiComm()):
         except Exception as e:
             print(f"Error checking CUDA context: {e}")
 
-        # Get MPI communicator
-        # comm = MpiComm()
-        # comm = McastDeviceMemory.get_comm()
-        # if config:
-        #     comm = config.comm_backend
-        # else:
-        #     comm = MpiComm()
-
         # Set up allocation properties
         handle_type = cuda.CUmemAllocationHandleType.CU_MEM_HANDLE_TYPE_FABRIC
 
@@ -838,9 +828,6 @@ def _alloc_mn_mcast_mem(self, buf_size: int, comm: Any=MpiComm()):
         )
 
         # All-gather fabric handles
-        print(my_fabric_handle.data)
-        print(type(my_fabric_handle.data))
-        # all_fabric_handles=[my_fabric_handle.data] * 4
         all_fabric_handles = comm.allgather(my_fabric_handle.data)
         cuda.cuCtxSynchronize()