[Platform] format platform to make it more clear (#610)

Platform should only contain the function that based from vllm. This PR
move the unrelated function to the right place to make platform more
clear.

Signed-off-by: wangxiyuan <wangxiyuan1007@gmail.com>

vllm_ascend/patch/__init__.py

@@ -62,7 +62,7 @@
 #       inside of the repo, and needs a common interface to destroy them, this patch add the interface of destroy
 #       platform owned `CoordinatorGroup` to make sure all the CoordinateGroup can be properly destroyed
 #    How：
-#       Call platform method `destroy_platform_model_parallel` to destroy all the `CoordinateGroup`
+#       Call `vllm_ascend.distributed.parallel_state method `destroy_platform_model_parallel` to destroy all the `CoordinateGroup`
 #    Related PR (if no, explain why): no related PR, we want add this ability into vllm
 #    Future Plan:
 #       Remove those patch when vllm merged them
@@ -73,7 +73,7 @@
 #       call to the `stateless_init_torch_distributed_process_group`, to enable other platform which may support
 #       stateless process group initialize method
 #    How：
-#       Call platform method `platform_has_backend_register` to judge if there is a stateless process group initialize
+#       rewrite stateless_init_torch_distributed_process_group to judge if there is a stateless process group initialize
 #       method and call platform method `platform_register_backend` to initialize them
 #    Related PR (if no, explain why): no related PR, we want add this ability into vllm
 #    Future Plan:

vllm_ascend/patch/platform/patch_common/patch_distributed.py

@@ -42,8 +42,9 @@ def ascend_destroy_model_parallel():
     if _DP:
         _DP.destroy()
     _DP = None
-    from vllm.platforms import current_platform
-    current_platform.destroy_platform_model_parallel()
+    from vllm_ascend.distributed.parallel_state import \
+        destory_ascend_model_parallel
+    destory_ascend_model_parallel()
 
 
 def ascend_stateless_init_torch_distributed_process_group(
@@ -100,7 +101,6 @@ def ascend_stateless_init_torch_distributed_process_group(
         group_rank,
         group_size,
     )
-    from vllm.platforms import current_platform
     if backend == "gloo":
         from torch.distributed.distributed_c10d import ProcessGroupGloo
         backend_class = ProcessGroupGloo(prefix_store,
@@ -120,8 +120,18 @@ def ascend_stateless_init_torch_distributed_process_group(
                                          backend_options)
         backend_type = ProcessGroup.BackendType.NCCL
         device = torch.device("cuda")
-    elif current_platform.platform_has_backend_register():
-        current_platform.platform_register_backend()
+    elif backend == "hccl":
+        from torch.distributed import is_hccl_available
+        assert is_hccl_available()
+        from torch_npu._C._distributed_c10d import ProcessGroupHCCL
+        backend_options = ProcessGroupHCCL.Options()
+        backend_options._timeout = timeout
+        backend_class = ProcessGroupHCCL(prefix_store, group_rank, group_size,
+                                         backend_options)
+        device = torch.device("npu")
+        backend_class._set_sequence_number_for_group()
+        backend_type = ProcessGroup.BackendType.CUSTOM
+        pg._register_backend(device, backend_type, backend_class)
         return pg
     else:
         raise RuntimeError(f"Unsupported torch distributed backend: {backend}")

vllm_ascend/platform.py

@@ -226,29 +226,3 @@ class NPUPlatform(Platform):
         model configuration.
         """
         return True
-
-    @classmethod
-    def destroy_platform_model_parallel(cls) -> None:
-        from vllm_ascend.distributed.parallel_state import \
-            destory_ascend_model_parallel
-        destory_ascend_model_parallel()
-
-    @classmethod
-    def platform_has_backend_register(cls) -> bool:
-        return True
-
-    @classmethod
-    def platform_register_backend(cls, pg, prefix_store, group_rank,
-                                  group_size, backend_options,
-                                  timeout) -> None:
-        from torch.distributed import ProcessGroup, is_hccl_available
-        assert is_hccl_available()
-        from torch_npu._C._distributed_c10d import ProcessGroupHCCL
-        backend_options = ProcessGroupHCCL.Options()
-        backend_options._timeout = timeout
-        backend_class = ProcessGroupHCCL(prefix_store, group_rank, group_size,
-                                         backend_options)
-        device = torch.device("npu")
-        backend_class._set_sequence_number_for_group()
-        backend_type = ProcessGroup.BackendType.CUSTOM
-        pg._register_backend(device, backend_type, backend_class)