Create torch_compile_conv_bn_fuser tutorial adapted from fx_conv_bn_fuser (#3458)

Author: yf225

.jenkins/validate_tutorials_built.py

@@ -23,7 +23,7 @@
     "beginner_source/examples_autograd/polynomial_autograd",
     "beginner_source/examples_autograd/polynomial_custom_function",
     "intermediate_source/mnist_train_nas",  # used by ax_multiobjective_nas_tutorial.py
-    "intermediate_source/fx_conv_bn_fuser",
+    "intermediate_source/torch_compile_conv_bn_fuser",
     "intermediate_source/_torch_export_nightly_tutorial",  # does not work on release
     "advanced_source/usb_semisup_learn", # fails with CUDA OOM error, should try on a different worker
     "prototype_source/fx_graph_mode_ptq_dynamic",

index.rst

@@ -348,13 +348,6 @@ Welcome to PyTorch Tutorials
 
 .. Code Transformations with FX
 
-.. customcarditem::
-   :header: Building a Convolution/Batch Norm fuser in FX
-   :card_description: Build a simple FX pass that fuses batch norm into convolution to improve performance during inference.
-   :image: _static/img/thumbnails/cropped/Deploying-PyTorch-in-Python-via-a-REST-API-with-Flask.png
-   :link: intermediate/fx_conv_bn_fuser.html
-   :tags: FX
-
 .. customcarditem::
    :header: Building a Simple Performance Profiler with FX
    :card_description: Build a simple FX interpreter to record the runtime of op, module, and function calls and report statistics
@@ -583,6 +576,13 @@ Welcome to PyTorch Tutorials
    :link: intermediate/torch_compile_tutorial.html
    :tags: Model-Optimization
 
+.. customcarditem::
+   :header: Building a Convolution/Batch Norm fuser in torch.compile
+   :card_description: Build a simple pattern matcher pass that fuses batch norm into convolution to improve performance during inference.
+   :image: _static/img/thumbnails/cropped/generic-pytorch-logo.png
+   :link: intermediate/torch_compile_conv_bn_fuser.html
+   :tags: Model-Optimization
+
 .. customcarditem::
    :header: Inductor CPU Backend Debugging and Profiling
    :card_description: Learn the usage, debugging and performance profiling for ``torch.compile`` with Inductor CPU backend.
@@ -950,7 +950,6 @@ Additional Resources
    :hidden:
    :caption: Code Transforms with FX
 
-   intermediate/fx_conv_bn_fuser
    intermediate/fx_profiling_tutorial
 
 .. toctree::
@@ -1001,6 +1000,7 @@ Additional Resources
    intermediate/nvfuser_intro_tutorial
    intermediate/ax_multiobjective_nas_tutorial
    intermediate/torch_compile_tutorial
+   intermediate/torch_compile_conv_bn_fuser
    intermediate/compiled_autograd_tutorial
    intermediate/inductor_debug_cpu
    intermediate/scaled_dot_product_attention_tutorial

intermediate_source/fx_conv_bn_fuser.py

@@ -216,9 +216,6 @@ def summary(self, should_sort : bool = False) -> str:
 #
 # * ``MaxPool2d`` takes up the most time. This is a known issue:
 #   https://github.com/pytorch/pytorch/issues/51393
-# * BatchNorm2d also takes up significant time. We can continue this
-#   line of thinking and optimize this in the Conv-BN Fusion with FX
-#   `tutorial <https://pytorch.org/tutorials/intermediate/fx_conv_bn_fuser.html>`_. 
 #
 #
 # Conclusion