From 81ec83c8099c1046fb52b706597ecb1e81794937 Mon Sep 17 00:00:00 2001
From: smishra8 <kumar0m@yahoo.com>
Date: Thu, 20 Nov 2025 08:49:13 -0500
Subject: [PATCH] Improve Quick Start tutorial for CPU-only Windows setup -
 draft

---
 docs/source/quick_start.rst | 35 +++++++++++++++++++++++++++++++++++
 1 file changed, 35 insertions(+)

diff --git a/docs/source/quick_start.rst b/docs/source/quick_start.rst
index 52947b7622..714f6bb6fe 100644
--- a/docs/source/quick_start.rst
+++ b/docs/source/quick_start.rst
@@ -16,6 +16,41 @@ need to install (`detailed instructions <https://pytorch.org/get-started/locally
 
   pip install torch
 
+.. note::
+
+   On Windows CPU-only builds (e.g. ``torch==2.9.0+cpu`` with ``torchao==0.14.0``),
+   TorchAO skips C++ extensions and advanced CUDA features. The following Int8
+   example compiles and runs successfully on CPU.
+
+.. code-block:: python
+
+   import torch
+   import torch.nn as nn
+   from torchao.quantization import Int8WeightOnlyConfig, quantize_
+
+   class TinyMLP(nn.Module):
+       def __init__(self):
+           super().__init__()
+           self.fc1 = nn.Linear(16, 32)
+           self.relu = nn.ReLU()
+           self.fc2 = nn.Linear(32, 8)
+
+       def forward(self, x):
+           return self.fc2(self.relu(self.fc1(x)))
+
+   model = TinyMLP().eval()
+   example_inputs = (torch.randn(1, 16),)
+
+   quantize_(model, Int8WeightOnlyConfig())
+
+   with torch.no_grad():
+       print("Output:", model(*example_inputs))
+
+This is a very basic, small example which is runnable on CPU-only builds. 
+We have a tiny MLP model which is two linear layers with a ReLU in between, and we have small input/output sizes which can be quickly run on CPU. 
+This toy model demonstrates int8 quantization which is supported on CPU, unlike int4 or bfloat16 which is better suited for GPU.
+Many new users may install PyTorch with +cpu by default or there may be a case where GPU is not available. 
+In this case, it may be helpful to run an example which is safe for CPUs, and it effectively shows that TorchAO works in a basic example.   
 
 First Quantization Example
 ==========================