[OpenReg] Fix the docs of Accelerator Intergration (#162826)

----

- Fixed the redirect link about step 1
- Formatted the autoload and added necessary links
Pull Request resolved: https://github.com/pytorch/pytorch/pull/162826
Approved by: https://github.com/albanD
ghstack dependencies: #161917, #161918, #160101

docs/source/accelerator/autoload.md

@@ -22,7 +22,7 @@ This tutorial will take **OpenReg** as a new out-of-the-tree device and guide yo
 
 ### Entry Point Setup
 
-To enable **Autoload**, register the `_autoload` function as an entry point in `setup.py` file.
+To enable **Autoload**, register the `_autoload` function as an entry point in [setup.py](https://github.com/pytorch/pytorch/blob/main/test/cpp_extensions/open_registration_extension/torch_openreg/setup.py) file.
 
 ::::{tab-set}
 
@@ -43,19 +43,18 @@ To enable **Autoload**, register the `_autoload` function as an entry point in `
 
 ### Backend Setup
 
-Define the initialization hook `_autoload` for backend initialization. This hook will be automatically invoked by PyTorch during startup.
+Define the initialization hook `_autoload` for backend initialization in [torch_openreg](https://github.com/pytorch/pytorch/blob/main/test/cpp_extensions/open_registration_extension/torch_openreg/torch_openreg/__init__.py). This hook will be automatically invoked by PyTorch during startup.
 
 ::::{tab-set-code}
+
 ```{eval-rst}
 .. literalinclude:: ../../../test/cpp_extensions/open_registration_extension/torch_openreg/torch_openreg/__init__.py
     :language: python
     :start-after: LITERALINCLUDE START: AUTOLOAD
     :end-before: LITERALINCLUDE END: AUTOLOAD
     :linenos:
-    :emphasize-lines: 10-12
 ```
 
-
 ::::
 
 ## Result
@@ -66,9 +65,6 @@ After setting up the entry point and backend, build and install your backend. No
 .. grid:: 2
 
     .. grid-item-card:: :octicon:`terminal;1em;` Without Autoload
-       :class-card: card-prerequisites
-
-       ::
 
            >>> import torch
            >>> import torch_openreg
@@ -76,11 +72,9 @@ After setting up the entry point and backend, build and install your backend. No
            tensor(1, device='openreg:0')
 
     .. grid-item-card:: :octicon:`terminal;1em;` With Autoload
-       :class-card: card-prerequisites
-
-       ::
 
            >>> import torch # Automatically import torch_openreg
+           >>>
            >>> torch.tensor(1, device="openreg")
            tensor(1, device='openreg:0')
 ```

docs/source/accelerator/operators.md

@@ -169,7 +169,7 @@ Of course, global fallbacks can also be combined with a blacklist of fallbacks,
 
 ### PyTorch STUB
 
-PyTorch also provides another approach for built-in operators: `STUB`. This method is essentially based on the `Step 1<step-one>` approach, but adds secondary scheduling capabilities (for example, scheduling based on CPU characteristics).
+PyTorch also provides another approach for built-in operators: `STUB`. This method is essentially based on the {ref}`Step 1<step-one>` approach, but adds secondary scheduling capabilities (for example, scheduling based on CPU characteristics).
 
 ```{note}
 The `STUB` method currently supports only a limited set of operators. For new accelerator devices, the advantage of the `STUB` method is that it significantly reduces the cost of development at the cost of a small performance overhead. PyTorch currently does not clearly list the set of operators that can be registered through `STUB`. Due to the large number of related operators, only the query method for the supported operator list is provided here.

test/cpp_extensions/open_registration_extension/torch_openreg/torch_openreg/__init__.py

@@ -9,7 +9,6 @@ if sys.platform == "win32":
     _load_dll_libraries()
     del _load_dll_libraries
 
-# LITERALINCLUDE START: AUTOLOAD
 import torch_openreg._C  # type: ignore[misc]
 import torch_openreg.openreg
 
@@ -19,6 +18,7 @@ torch._register_device_module("openreg", torch_openreg.openreg)
 torch.utils.generate_methods_for_privateuse1_backend(for_storage=True)
 
 
+# LITERALINCLUDE START: AUTOLOAD
 def _autoload():
     # It is a placeholder function here to be registered as an entry point.
     pass