Merge branch 'main' into readme-update

Author: svekars

advanced_source/cpp_custom_ops.rst

@@ -174,6 +174,8 @@ To add ``torch.compile`` support for an operator, we must add a FakeTensor kerne
 known as a "meta kernel" or "abstract impl"). FakeTensors are Tensors that have
 metadata (such as shape, dtype, device) but no data: the FakeTensor kernel for an
 operator specifies how to compute the metadata of output tensors given the metadata of input tensors.
+The FakeTensor kernel should return dummy Tensors of your choice with
+the correct Tensor metadata (shape/strides/``dtype``/device).
 
 We recommend that this be done from Python via the `torch.library.register_fake` API,
 though it is possible to do this from C++ as well (see

advanced_source/dynamic_quantization_tutorial.py

@@ -151,7 +151,8 @@ def tokenize(self, path):
 model.load_state_dict(
     torch.load(
         model_data_filepath + 'word_language_model_quantize.pth',
-        map_location=torch.device('cpu')
+        map_location=torch.device('cpu'),
+        weights_only=True
         )
     )
 

advanced_source/python_custom_ops.py

@@ -66,7 +66,7 @@ def display(img):
 ######################################################################
 # ``crop`` is not handled effectively out-of-the-box by
 # ``torch.compile``: ``torch.compile`` induces a
-# `"graph break" <https://pytorch.org/docs/stable/torch.compiler_faq.html#graph-breaks>`_ 
+# `"graph break" <https://pytorch.org/docs/stable/torch.compiler_faq.html#graph-breaks>`_
 # on functions it is unable to handle and graph breaks are bad for performance.
 # The following code demonstrates this by raising an error
 # (``torch.compile`` with ``fullgraph=True`` raises an error if a
@@ -85,9 +85,9 @@ def f(img):
 #
 # 1. wrap the function into a PyTorch custom operator.
 # 2. add a "``FakeTensor`` kernel" (aka "meta kernel") to the operator.
-#    Given the metadata (e.g. shapes)
-#    of the input Tensors, this function says how to compute the metadata
-#    of the output Tensor(s).
+#    Given some ``FakeTensors`` inputs (dummy Tensors that don't have storage),
+#    this function should return dummy Tensors of your choice with the correct
+#    Tensor metadata (shape/strides/``dtype``/device).
 
 
 from typing import Sequence
@@ -130,6 +130,11 @@ def f(img):
 # ``autograd.Function`` with PyTorch operator registration APIs can lead to (and
 # has led to) silent incorrectness when composed with ``torch.compile``.
 #
+# If you don't need training support, there is no need to use
+# ``torch.library.register_autograd``.
+# If you end up training with a ``custom_op`` that doesn't have an autograd
+# registration, we'll raise an error message.
+#
 # The gradient formula for ``crop`` is essentially ``PIL.paste`` (we'll leave the
 # derivation as an exercise to the reader). Let's first wrap ``paste`` into a
 # custom operator:
@@ -203,7 +208,7 @@ def setup_context(ctx, inputs, output):
 ######################################################################
 # Mutable Python Custom operators
 # -------------------------------
-# You can also wrap a Python function that mutates its inputs into a custom 
+# You can also wrap a Python function that mutates its inputs into a custom
 # operator.
 # Functions that mutate inputs are common because that is how many low-level
 # kernels are written; for example, a kernel that computes ``sin`` may take in

advanced_source/static_quantization_tutorial.rst

@@ -286,7 +286,7 @@ We next define several helper functions to help with model evaluation. These mos
 
     def load_model(model_file): 
         model = MobileNetV2() 
-        state_dict = torch.load(model_file) 
+        state_dict = torch.load(model_file, weights_only=True) 
         model.load_state_dict(state_dict) 
         model.to('cpu') 
         return model  

beginner_source/basics/quickstart_tutorial.py

@@ -216,7 +216,7 @@ def test(dataloader, model, loss_fn):
 # the state dictionary into it.
 
 model = NeuralNetwork().to(device)
-model.load_state_dict(torch.load("model.pth"))
+model.load_state_dict(torch.load("model.pth", weights_only=True))
 
 #############################################################
 # This model can now be used to make predictions.

beginner_source/basics/saveloadrun_tutorial.py

@@ -32,9 +32,14 @@
 ##########################
 # To load model weights, you need to create an instance of the same model first, and then load the parameters
 # using ``load_state_dict()`` method.
+#
+# In the code below, we set ``weights_only=True`` to limit the
+# functions executed during unpickling to only those necessary for
+# loading weights. Using ``weights_only=True`` is considered
+# a best practice when loading weights.
 
 model = models.vgg16() # we do not specify ``weights``, i.e. create untrained model
-model.load_state_dict(torch.load('model_weights.pth'))
+model.load_state_dict(torch.load('model_weights.pth', weights_only=True))
 model.eval()
 
 ###########################
@@ -50,9 +55,14 @@
 torch.save(model, 'model.pth')
 
 ########################
-# We can then load the model like this:
+# We can then load the model as demonstrated below.
+#
+# As described in `Saving and loading torch.nn.Modules <pytorch.org/docs/main/notes/serialization.html#saving-and-loading-torch-nn-modules>`__,
+# saving ``state_dict``s is considered the best practice. However,
+# below we use ``weights_only=False`` because this involves loading the
+# model, which is a legacy use case for ``torch.save``.
 
-model = torch.load('model.pth')
+model = torch.load('model.pth', weights_only=False),
 
 ########################
 # .. note:: This approach uses Python `pickle <https://docs.python.org/3/library/pickle.html>`_ module when serializing the model, thus it relies on the actual class definition to be available when loading the model.

beginner_source/blitz/cifar10_tutorial.py

@@ -221,7 +221,7 @@ def forward(self, x):
 # wasn't necessary here, we only did it to illustrate how to do so):
 
 net = Net()
-net.load_state_dict(torch.load(PATH))
+net.load_state_dict(torch.load(PATH, weights_only=True))
 
 ########################################################################
 # Okay, now let us see what the neural network thinks these examples above are:

beginner_source/fgsm_tutorial.py

@@ -192,7 +192,7 @@ def forward(self, x):
 model = Net().to(device)
 
 # Load the pretrained model
-model.load_state_dict(torch.load(pretrained_model, map_location=device))
+model.load_state_dict(torch.load(pretrained_model, map_location=device, weights_only=True))
 
 # Set the model in evaluation mode. In this case this is for the Dropout layers
 model.eval()

beginner_source/saving_loading_models.py

@@ -153,7 +153,7 @@
 # .. code:: python
 #
 #    model = TheModelClass(*args, **kwargs)
-#    model.load_state_dict(torch.load(PATH))
+#    model.load_state_dict(torch.load(PATH), weights_only=True)
 #    model.eval()
 #
 # .. note::
@@ -206,7 +206,7 @@
 # .. code:: python
 #
 #    # Model class must be defined somewhere
-#    model = torch.load(PATH)
+#    model = torch.load(PATH, weights_only=False)
 #    model.eval()
 #
 # This save/load process uses the most intuitive syntax and involves the
@@ -290,7 +290,7 @@
 #    model = TheModelClass(*args, **kwargs)
 #    optimizer = TheOptimizerClass(*args, **kwargs)
 #
-#    checkpoint = torch.load(PATH)
+#    checkpoint = torch.load(PATH, weights_only=True)
 #    model.load_state_dict(checkpoint['model_state_dict'])
 #    optimizer.load_state_dict(checkpoint['optimizer_state_dict'])
 #    epoch = checkpoint['epoch']
@@ -354,7 +354,7 @@
 #    optimizerA = TheOptimizerAClass(*args, **kwargs)
 #    optimizerB = TheOptimizerBClass(*args, **kwargs)
 #
-#    checkpoint = torch.load(PATH)
+#    checkpoint = torch.load(PATH, weights_only=True)
 #    modelA.load_state_dict(checkpoint['modelA_state_dict'])
 #    modelB.load_state_dict(checkpoint['modelB_state_dict'])
 #    optimizerA.load_state_dict(checkpoint['optimizerA_state_dict'])
@@ -407,7 +407,7 @@
 # .. code:: python
 #
 #    modelB = TheModelBClass(*args, **kwargs)
-#    modelB.load_state_dict(torch.load(PATH), strict=False)
+#    modelB.load_state_dict(torch.load(PATH), strict=False, weights_only=True)
 #
 # Partially loading a model or loading a partial model are common
 # scenarios when transfer learning or training a new complex model.
@@ -446,7 +446,7 @@
 #
 #    device = torch.device('cpu')
 #    model = TheModelClass(*args, **kwargs)
-#    model.load_state_dict(torch.load(PATH, map_location=device))
+#    model.load_state_dict(torch.load(PATH, map_location=device, weights_only=True))
 #
 # When loading a model on a CPU that was trained with a GPU, pass
 # ``torch.device('cpu')`` to the ``map_location`` argument in the
@@ -469,7 +469,7 @@
 #
 #    device = torch.device("cuda")
 #    model = TheModelClass(*args, **kwargs)
-#    model.load_state_dict(torch.load(PATH))
+#    model.load_state_dict(torch.load(PATH, weights_only=True))
 #    model.to(device)
 #    # Make sure to call input = input.to(device) on any input tensors that you feed to the model
 #
@@ -497,7 +497,7 @@
 #
 #    device = torch.device("cuda")
 #    model = TheModelClass(*args, **kwargs)
-#    model.load_state_dict(torch.load(PATH, map_location="cuda:0"))  # Choose whatever GPU device number you want
+#    model.load_state_dict(torch.load(PATH, weights_only=True, map_location="cuda:0"))  # Choose whatever GPU device number you want
 #    model.to(device)
 #    # Make sure to call input = input.to(device) on any input tensors that you feed to the model
 #

beginner_source/transfer_learning_tutorial.py

@@ -209,7 +209,7 @@ def train_model(model, criterion, optimizer, scheduler, num_epochs=25):
         print(f'Best val Acc: {best_acc:4f}')
 
         # load best model weights
-        model.load_state_dict(torch.load(best_model_params_path))
+        model.load_state_dict(torch.load(best_model_params_path, weights_only=True))
     return model