Fix multiple typing issues

Author: gruebel

ignite/distributed/auto.py

@@ -120,7 +120,7 @@ def auto_dataloader(dataset: Dataset, **kwargs: Any) -> Union[DataLoader, "_MpDe
 
         sampler = dataloader.sampler  # type: ignore[union-attr]
         dataloader = mp_device_loader_cls(dataloader, idist.device())
-        dataloader.sampler = sampler
+        dataloader.sampler = sampler  # type: ignore[attr-defined]
 
     return dataloader
 
@@ -320,12 +320,6 @@ def __init__(self, loader: Any, device: torch.device, **kwargs: Any) -> None:
             self._device = device
             self._parallel_loader_kwargs = kwargs
 
-        def __setattr__(self, name: str, value: Any) -> None:
-            super().__setattr__(name, value)
-
-        def __getattr__(self, name: str) -> Any:
-            super().__getattribute__(name)
-
         def __iter__(self) -> Iterator:
             parallel_loader = ParallelLoader(self._loader, [self._device], **self._parallel_loader_kwargs)
             return parallel_loader.per_device_loader(self._device)

ignite/distributed/comp_models/base.py

@@ -221,7 +221,7 @@ class _SerialModel(ComputationModel):
     name = "serial"
     available_backends = ()
 
-    def __init__(self, _backend: Optional[str] = None, **_kwargs: Any) -> None:
+    def __init__(self, _backend: Optional[str] = None, **kwargs: Any) -> None:
         super(_SerialModel, self).__init__()
 
     def get_local_rank(self) -> int:
@@ -271,8 +271,8 @@ def spawn(*args: Any, **kwargs: Any) -> None:
     def all_reduce(self, tensor: Union[torch.Tensor, Number], op: str = "sum") -> Union[torch.Tensor, Number]:
         return tensor
 
-    def all_gather(self, tensor: Union[torch.Tensor, Number, str]) -> Union[torch.Tensor, Number, List[str]]:
-        return cast(Union[torch.Tensor, Number], tensor)
+    def all_gather(self, tensor: Union[torch.Tensor, Number]) -> Union[torch.Tensor, Number]:  # type: ignore
+        return tensor
 
     def broadcast(self, tensor: Union[torch.Tensor, Number, str], src: int = 0) -> Union[torch.Tensor, Number, str]:
         return tensor

ignite/distributed/comp_models/native.py

@@ -63,7 +63,7 @@ def __init__(self, backend: Optional[str] = None, timeout: Optional[int] = None,
             else:
                 self._init_from_context()
 
-        def _create_from_backend(self, backend: str, timeout: Optional[int] = None, **_kwargs: Any) -> None:
+        def _create_from_backend(self, backend: str, timeout: Optional[int] = None, **kwargs: Any) -> None:
             if backend == dist.Backend.NCCL and not torch.cuda.is_available():
                 raise RuntimeError("Nccl backend is required but no cuda capable devices")
 

ignite/distributed/comp_models/xla.py

@@ -50,7 +50,7 @@ def __init__(self, backend: Optional[str] = None, **kwargs: Any):
             else:
                 self._init_from_context()
 
-        def _create_from_backend(self, backend: str, **_kwargs: Any) -> None:
+        def _create_from_backend(self, backend: str, **kwargs: Any) -> None:
             xm.rendezvous("init")
 
             self._backend = backend

ignite/distributed/launcher.py

@@ -216,10 +216,10 @@ def __init__(
     @staticmethod
     def _setup_spawn_params(
         nproc_per_node: int,
-        nnodes: Optional[int],
-        node_rank: Optional[int],
-        master_addr: Optional[str],
-        master_port: Optional[int],
+        nnodes: Optional[int] = None,
+        node_rank: Optional[int] = None,
+        master_addr: Optional[str] = None,
+        master_port: Optional[int] = None,
         **spawn_kwargs: Any
     ) -> Dict:
         if nproc_per_node < 1:
@@ -273,7 +273,7 @@ def training(local_rank, config, **kwargs):
             **kwargs: keyword arguments of ``func``.
 
         """
-        if self._spawn_params is not None:
+        if self._spawn_params is not None and self.backend is not None:
             self.logger.info("Spawn function '{}' in {} processes".format(func, self._spawn_params["nproc_per_node"]))
             idist.spawn(self.backend, func, args=args, kwargs_dict=kwargs, **self._spawn_params)
         else:

ignite/distributed/utils.py

@@ -190,7 +190,7 @@ def hostname() -> str:
 
 
 def spawn(
-    backend: Optional[str],
+    backend: str,
     fn: Callable,
     args: Tuple,
     kwargs_dict: Optional[Mapping] = None,
@@ -349,7 +349,7 @@ def all_gather(tensor: Union[torch.Tensor, Number, str]) -> Union[torch.Tensor,
     if _need_to_sync and isinstance(_model, _SerialModel):
         sync(temporary=True)
 
-    return _model.all_gather(tensor)
+    return _model.all_gather(tensor)  # type: ignore[arg-type]
 
 
 def broadcast(tensor: Union[torch.Tensor, Number, str], src: int = 0) -> Union[torch.Tensor, Number, str]:
@@ -440,7 +440,7 @@ def _set_model(model: Any, temporary: bool = False) -> None:
         _need_to_sync = False
 
 
-def _assert_backend(backend: Optional[str]) -> None:
+def _assert_backend(backend: str) -> None:
     backends = available_backends()
     if backend not in backends:
         raise ValueError("Backend should be one of '{}'".format(backends))
@@ -527,7 +527,7 @@ def show_config() -> None:
     logger.info("node rank: {}".format(get_node_rank()))
 
 
-def one_rank_only(rank: int = 0, with_barrier: bool = False) -> Optional[Callable]:
+def one_rank_only(rank: int = 0, with_barrier: bool = False) -> Callable:
     """Decorator to filter handlers wrt a rank number
 
     Args:
@@ -549,9 +549,9 @@ def some_handler(_):
             ...
     """
 
-    def _one_rank_only(func: Callable) -> Optional[Callable]:
+    def _one_rank_only(func: Callable) -> Callable:
         @wraps(func)
-        def wrapper(*args: Any, **kwargs: Any) -> Optional[Callable]:
+        def wrapper(*args: Any, **kwargs: Any) -> Optional[Any]:
             ret = None
             if get_rank() == rank:
                 ret = func(*args, **kwargs)