Merge branch 'main' into swap

Author: svekars

.ci/docker/Dockerfile

@@ -7,6 +7,11 @@ ENV DEBIAN_FRONTEND noninteractive
 COPY ./common/install_base.sh install_base.sh
 RUN bash ./install_base.sh && rm install_base.sh
 
+# Setup user
+# TODO: figure out how to remove this part
+COPY ./common/install_user.sh install_user.sh
+RUN bash ./install_user.sh && rm install_user.sh
+
 COPY ./common/install_docs_reqs.sh install_docs_reqs.sh
 RUN bash ./install_docs_reqs.sh && rm install_docs_reqs.sh
 
@@ -20,4 +25,5 @@ COPY ./common/install_conda.sh install_conda.sh
 COPY ./common/common_utils.sh common_utils.sh
 RUN bash ./install_conda.sh && rm install_conda.sh common_utils.sh /opt/conda/requirements.txt
 
+USER ci-user
 CMD ["bash"]

.ci/docker/common/common_utils.sh

@@ -7,7 +7,7 @@ as_ci_user() {
   # NB: Pass on PATH and LD_LIBRARY_PATH to sudo invocation
   # NB: This must be run from a directory that the user has access to,
   # works around https://github.com/conda/conda-package-handling/pull/34
-  sudo -E -H env -u SUDO_UID -u SUDO_GID -u SUDO_COMMAND -u SUDO_USER env "PATH=$PATH" "LD_LIBRARY_PATH=$LD_LIBRARY_PATH" $*
+  sudo -E -H -u ci-user env -u SUDO_UID -u SUDO_GID -u SUDO_COMMAND -u SUDO_USER env "PATH=$PATH" "LD_LIBRARY_PATH=$LD_LIBRARY_PATH" $*
 }
 
 conda_install() {

.ci/docker/common/install_conda.sh

@@ -12,6 +12,7 @@ if [ -n "$ANACONDA_PYTHON_VERSION" ]; then
   CONDA_FILE="Miniconda3-latest-Linux-x86_64.sh"
 
   mkdir -p /opt/conda
+  chown ci-user:ci-user /opt/conda
 
   source "$(dirname "${BASH_SOURCE[0]}")/common_utils.sh"
 

.ci/docker/common/install_user.sh

@@ -0,0 +1,26 @@
+#!/bin/bash
+# Copyright (c) Meta Platforms, Inc. and affiliates.
+# All rights reserved.
+#
+# This source code is licensed under the BSD-style license found in the
+# LICENSE file in the root directory of this source tree.
+
+# Copied from https://github.com/pytorch/executorch/blob/6e431355a554e5f84c3a05dfa2b981ead90c2b48/.ci/docker/common/install_user.sh#L1
+
+set -ex
+
+# Same as ec2-user
+echo "ci-user:x:1000:1000::/var/lib/ci-user:" >> /etc/passwd
+echo "ci-user:x:1000:" >> /etc/group
+# Needed on Focal or newer
+echo "ci-user:*:19110:0:99999:7:::" >> /etc/shadow
+
+# Create $HOME
+mkdir -p /var/lib/ci-user
+chown ci-user:ci-user /var/lib/ci-user
+
+# Allow sudo
+echo 'ci-user ALL=(ALL) NOPASSWD:ALL' > /etc/sudoers.d/ci-user
+
+# Test that sudo works
+sudo -u ci-user sudo -v

.ci/docker/requirements.txt

@@ -4,7 +4,6 @@
 sphinx==5.0.0
 sphinx-gallery==0.11.1
 sphinx_design
-nbsphinx
 docutils==0.16
 sphinx-copybutton
 pypandoc==1.12

.github/workflows/build-tutorials.yml

@@ -93,7 +93,7 @@ jobs:
             "${DOCKER_IMAGE}"
           )
 
-          docker exec -t "${container_name}" sh -c ".jenkins/build.sh"
+          docker exec -u ci-user -t "${container_name}" sh -c ".jenkins/build.sh"
 
       - name: Teardown Linux
         uses: pytorch/test-infra/.github/actions/teardown-linux@main
@@ -162,9 +162,7 @@ jobs:
             "${DOCKER_IMAGE}"
           )
 
-          docker exec -u root -i "${container_name}" bash
-
-          docker exec -t "${container_name}" sh -c ".jenkins/build.sh"
+          docker exec -u ci-user -t "${container_name}" sh -c ".jenkins/build.sh"
 
       - name: Upload docs preview
         uses: seemethere/upload-artifact-s3@v5

beginner_source/blitz/neural_networks_tutorial.py

@@ -55,16 +55,33 @@ def __init__(self):
         self.fc2 = nn.Linear(120, 84)
         self.fc3 = nn.Linear(84, 10)
 
-    def forward(self, x):
-        # Max pooling over a (2, 2) window
-        x = F.max_pool2d(F.relu(self.conv1(x)), (2, 2))
-        # If the size is a square, you can specify with a single number
-        x = F.max_pool2d(F.relu(self.conv2(x)), 2)
-        x = torch.flatten(x, 1) # flatten all dimensions except the batch dimension
-        x = F.relu(self.fc1(x))
-        x = F.relu(self.fc2(x))
-        x = self.fc3(x)
-        return x
+    def forward(self, input):
+        # Convolution layer C1: 1 input image channel, 6 output channels,
+        # 5x5 square convolution, it uses RELU activation function, and
+        # outputs a Tensor with size (N, 6, 28, 28), where N is the size of the batch
+        c1 = F.relu(self.conv1(input))
+        # Subsampling layer S2: 2x2 grid, purely functional,
+        # this layer does not have any parameter, and outputs a (N, 16, 14, 14) Tensor
+        s2 = F.max_pool2d(c1, (2, 2))
+        # Convolution layer C3: 6 input channels, 16 output channels,
+        # 5x5 square convolution, it uses RELU activation function, and
+        # outputs a (N, 16, 10, 10) Tensor
+        c3 = F.relu(self.conv2(s2))
+        # Subsampling layer S4: 2x2 grid, purely functional,
+        # this layer does not have any parameter, and outputs a (N, 16, 5, 5) Tensor
+        s4 = F.max_pool2d(c3, 2)
+        # Flatten operation: purely functional, outputs a (N, 400) Tensor
+        s4 = torch.flatten(s4, 1)
+        # Fully connected layer F5: (N, 400) Tensor input,
+        # and outputs a (N, 120) Tensor, it uses RELU activation function
+        f5 = F.relu(self.fc1(s4))
+        # Fully connected layer F6: (N, 120) Tensor input,
+        # and outputs a (N, 84) Tensor, it uses RELU activation function
+        f6 = F.relu(self.fc2(f5))
+        # Gaussian layer OUTPUT: (N, 84) Tensor input, and
+        # outputs a (N, 10) Tensor
+        output = self.fc3(f6)
+        return output
 
 
 net = Net()

conf.py

@@ -76,7 +76,6 @@
     'sphinx_copybutton',
     'sphinx_gallery.gen_gallery',
     'sphinx_design',
-    'nbsphinx'
 ]
 
 intersphinx_mapping = {

intermediate_source/FSDP_tutorial.rst

@@ -10,7 +10,7 @@ Training AI models at a large scale is a challenging task that requires a lot of
 It also comes with considerable engineering complexity to handle the training of these very large models.
 `PyTorch FSDP <https://pytorch.org/blog/introducing-pytorch-fully-sharded-data-parallel-api/>`__, released in PyTorch 1.11 makes this easier.
 
-In this tutorial, we show how to use `FSDP APIs <https://pytorch.org/docs/1.11/fsdp.html>`__, for simple MNIST models that can be extended to other larger models such as `HuggingFace BERT models <https://huggingface.co/blog/zero-deepspeed-fairscale>`__, 
+In this tutorial, we show how to use `FSDP APIs <https://pytorch.org/docs/stable/fsdp.html>`__, for simple MNIST models that can be extended to other larger models such as `HuggingFace BERT models <https://huggingface.co/blog/zero-deepspeed-fairscale>`__, 
 `GPT 3 models up to 1T parameters <https://pytorch.medium.com/training-a-1-trillion-parameter-model-with-pytorch-fully-sharded-data-parallel-on-aws-3ac13aa96cff>`__ . The sample DDP MNIST code has been borrowed from `here <https://github.com/yqhu/mnist_examples>`__. 
 
 
@@ -63,9 +63,7 @@ Here we use a toy model to run training on the MNIST dataset for demonstration p
 
 1.1 Install PyTorch along with Torchvision
 
-.. code-block:: bash 
-
-    pip3 install --pre torch torchvision torchaudio -f https://download.pytorch.org/whl/nightly/cu113/torch_nightly.html
+See the `Get Started guide <https://pytorch.org/get-started/locally/>`__ for information on installation.
 
 We add the following code snippets to a python script “FSDP_mnist.py”.