Fix torch.compile links (#121824)

Fixes https://github.com/pytorch/pytorch.github.io/issues/1567

Pull Request resolved: https://github.com/pytorch/pytorch/pull/121824
Approved by: https://github.com/svekars, https://github.com/peterbell10, https://github.com/malfet
ghstack dependencies: #121823

.github/ISSUE_TEMPLATE/pt2-bug-report.yml

@@ -33,7 +33,7 @@ body:
       label: Minified repro
       description: |
         Please run the minifier on your example and paste the minified code below
-        Learn more here https://pytorch.org/docs/main/compile/troubleshooting.html
+        Learn more here https://pytorch.org/docs/main/torch.compiler_troubleshooting.html
       placeholder: |
         env TORCHDYNAMO_REPRO_AFTER="aot" python your_model.py
         or

docs/source/torch.compiler_custom_backends.rst

@@ -84,7 +84,7 @@ Registration serves two purposes:
 
 * You can pass a string containing your backend function's name to ``torch.compile`` instead of the function itself,
   for example, ``torch.compile(model, backend="my_compiler")``.
-* It is required for use with the `minifier <https://pytorch.org/docs/main/compile/troubleshooting.html>`__. Any generated
+* It is required for use with the `minifier <https://pytorch.org/docs/main/torch.compiler_troubleshooting.html>`__. Any generated
   code from the minifier must call your code that registers your backend function, typically through an ``import`` statement.
 
 Custom Backends after AOTAutograd

docs/source/torch.rst

@@ -742,7 +742,7 @@ Optimizations
 
     compile
 
-`torch.compile documentation <https://pytorch.org/docs/main/compile/index.html>`__
+`torch.compile documentation <https://pytorch.org/docs/main/torch.compiler.html>`__
 
 Operator Tags
 ------------------------------------

torch/__init__.py

@@ -1817,7 +1817,7 @@ def compile(model: Optional[Callable] = None, *,
 
         - Experimental or debug in-tree backends can be seen with `torch._dynamo.list_backends(None)`
 
-        - To register an out-of-tree custom backend: https://pytorch.org/docs/main/compile/custom-backends.html
+        - To register an out-of-tree custom backend: https://pytorch.org/docs/main/torch.compiler_custom_backends.html
        mode (str): Can be either "default", "reduce-overhead", "max-autotune" or "max-autotune-no-cudagraphs"
 
         - "default" is the default mode, which is a good balance between performance and overhead

torch/_dynamo/utils.py

@@ -102,8 +102,10 @@ from torch.utils._triton import has_triton, has_triton_package
 
 counters: DefaultDict[str, Counter[str]] = collections.defaultdict(collections.Counter)
 optimus_scuba_log: Dict[str, Any] = {}
-troubleshooting_url = "https://pytorch.org/docs/main/compile/troubleshooting.html"
-nnmodule_doc_url = "https://pytorch.org/docs/main/compile/nn-module.html"
+troubleshooting_url = (
+    "https://pytorch.org/docs/main/torch.compiler_troubleshooting.html"
+)
+nnmodule_doc_url = "https://pytorch.org/docs/main/torch.compiler_nn_module.html"
 nnmodule_doc_url_msg = f"See {nnmodule_doc_url} for more information and limitations."
 log = logging.getLogger(__name__)
 

torch/fx/README.md

@@ -103,7 +103,7 @@ During the call to `symbolic_trace`, the parameter `x` is transformed into a Pro
 
 If you're doing graph transforms, you can wrap your own Proxy method around a raw Node so that you can use the overloaded operators to add additional things to a Graph.
 
-## [TorchDynamo](https://pytorch.org/docs/main/compile/technical-overview.html) ##
+## [TorchDynamo](https://pytorch.org/docs/main/torch.compiler_deepdive.html) ##
 
 Tracing has limitations in that it can't deal with dynamic control flow and is limited to outputting a single graph at a time, so a better alternative is the new `torch.compile()` infrastructure where you can output multiple subgraphs in either an aten or torch IR using `torch.fx`. [This tutorial](https://colab.research.google.com/drive/1Zh-Uo3TcTH8yYJF-LLo5rjlHVMtqvMdf) gives more context on how this works.