Convert directory fbcode/deeplearning to use the Ruff Formatter

benchmarks/models.py

@@ -8,7 +8,6 @@
 Contains models used for benchmarking
 """
 
-
 from dataclasses import dataclass
 from typing import Any
 

benchmarks/profiler.py

@@ -10,7 +10,6 @@
 Run via Jupyter
 """
 
-
 from benchmark import ModelBenchmarks
 
 

crypten/communicator/communicator.py

@@ -162,7 +162,6 @@ def _logging(func):
 
     @wraps(func)
     def logging_wrapper(self, *args, **kwargs):
-
         # TODO: Replace this
         # - hacks the inputs into some of the functions for world_size 1:
         world_size = self.get_world_size()

crypten/communicator/in_process_communicator.py

@@ -17,7 +17,6 @@
 
 
 class InProcessCommunicator(Communicator):
-
     BYTES_PER_ELEMENT = 8
     tls = threading.local()
     mailbox = None

crypten/cryptensor.py

@@ -78,16 +78,12 @@ def register_cryptensor_cls(cls):
             if name in CrypTensor.__CRYPTENSOR_TYPES__:
                 raise ValueError(
                     "Cannot register duplicate CrypTensor type: \
-                    tensor type {} already exists.".format(
-                        name
-                    )
+                    tensor type {} already exists.".format(name)
                 )
             if not issubclass(cls, CrypTensor):
                 raise ValueError(
                     "Registered tensor ({}: {}) must extend \
-                    CrypTensor".format(
-                        name, cls.__name__
-                    )
+                    CrypTensor".format(name, cls.__name__)
                 )
             CrypTensor.__CRYPTENSOR_TYPES__[name] = cls
             return cls
@@ -222,7 +218,6 @@ def backward(self, grad_input=None, top_node=True):
         """
         if self.requires_grad:
             with CrypTensor.no_grad():  # disable autograd for backward pass
-
                 # in initial backward call, identify all required nodes:
                 if top_node:
                     self._identify_required_grads()
@@ -302,7 +297,6 @@ def __torch_function__(self, func, types, args=(), kwargs=None):
             )
 
     def _get_forward_function_no_ctx(self, grad_fn):
-
         # determine if self is a dummy object (the case for staticmethods):
         is_dummy = getattr(self, "__IS_DUMMY__", False)
 
@@ -320,7 +314,6 @@ def autograd_forward_no_ctx(*args, **kwargs):
         return autograd_forward_no_ctx
 
     def _get_autograd_forward_function(self, name, grad_fn, in_place):
-
         # determine if self is a dummy object (the case for staticmethods):
         is_dummy = getattr(self, "__IS_DUMMY__", False)
 

crypten/gradients.py

@@ -408,7 +408,6 @@ def backward(ctx, grad_output):
 class AutogradSqueeze(AutogradFunction):
     @staticmethod
     def forward(ctx, *args, **kwargs):
-
         # preprocess inputs:
         assert len(args) >= 1
         if len(args) == 1:
@@ -497,7 +496,6 @@ def backward(ctx, grad_output):
 class AutogradDropout(AutogradFunction):
     @staticmethod
     def forward(ctx, input, p=0.5, training=True, inplace=False):
-
         if training and inplace:
             logging.warning(
                 "CrypTen dropout does not support inplace computation during training."
@@ -534,7 +532,6 @@ def backward(ctx, grad_output):
 class AutogradFeatureDropout(AutogradFunction):
     @staticmethod
     def forward(ctx, input, p=0.5, training=True, inplace=False):
-
         if training and inplace:
             logging.warning(
                 "CrypTen _feature_dropout does not support inplace computation during training."
@@ -1133,7 +1130,6 @@ def backward(ctx, grad_output):
 class AutogradSum(AutogradFunction):
     @staticmethod
     def forward(ctx, *args, **kwargs):
-
         # preprocess inputs:
         assert len(args) >= 1
         if len(args) == 1:
@@ -1192,7 +1188,6 @@ def backward(ctx, grad_output):
 class AutogradMean(AutogradFunction):
     @staticmethod
     def forward(ctx, *args, **kwargs):
-
         # preprocess inputs:
         assert len(args) >= 1
         if len(args) == 1:
@@ -1230,7 +1225,6 @@ def backward(ctx, grad_output):
 class AutogradVariance(AutogradFunction):
     @staticmethod
     def forward(ctx, self, *args, **kwargs):
-
         # preprocess inputs:
         if len(args) == 0:
             dim = None
@@ -1287,7 +1281,6 @@ def backward(ctx, grad_output):
 class AutogradMin(AutogradFunction):
     @staticmethod
     def forward(ctx, *args, **kwargs):
-
         # preprocess inputs:
         assert len(args) >= 1
         if len(args) == 1:
@@ -1335,7 +1328,6 @@ def backward(ctx, grad_output):
 class AutogradMax(AutogradFunction):
     @staticmethod
     def forward(ctx, *args, **kwargs):
-
         # preprocess inputs:
         assert len(args) >= 1
         if len(args) == 1:
@@ -1453,7 +1445,6 @@ def backward(ctx, grad_output):
 class AutogradAvgPool2D(AutogradFunction):
     @staticmethod
     def forward(ctx, input, kernel_size, stride=None, padding=0, ceil_mode=False):
-
         # preprocess inputs:
         if stride is None:
             stride = kernel_size
@@ -1528,7 +1519,6 @@ def forward(
         ceil_mode=False,
         return_indices=False,
     ):
-
         # preprocess inputs:
         if stride is None:
             stride = kernel_size
@@ -1887,7 +1877,6 @@ def backward(ctx, grad_output):
         grad_output = grad_output.mul(weight)
         grad_input = grad_output.mul(inv_var)
         if training:
-
             # compute gradient term that is due to the mean:
             num_element = reduce(
                 lambda x, y: x * y, [grad_output.size(d) for d in stats_dimensions]

crypten/mpc/primitives/arithmetic.py

@@ -223,9 +223,9 @@ def __setitem__(self, index, value):
         """Set tensor values by index"""
         if isinstance(value, (int, float)) or is_tensor(value):
             value = ArithmeticSharedTensor(value)
-        assert isinstance(
-            value, ArithmeticSharedTensor
-        ), "Unsupported input type %s for __setitem__" % type(value)
+        assert isinstance(value, ArithmeticSharedTensor), (
+            "Unsupported input type %s for __setitem__" % type(value)
+        )
         self.share.__setitem__(index, value.share)
 
     def pad(self, pad, mode="constant", value=0):
@@ -268,9 +268,9 @@ def stack(tensors, *args, **kwargs):
         for i, tensor in enumerate(tensors):
             if is_tensor(tensor):
                 tensors[i] = ArithmeticSharedTensor(tensor)
-            assert isinstance(
-                tensors[i], ArithmeticSharedTensor
-            ), "Can't stack %s with ArithmeticSharedTensor" % type(tensor)
+            assert isinstance(tensors[i], ArithmeticSharedTensor), (
+                "Can't stack %s with ArithmeticSharedTensor" % type(tensor)
+            )
 
         result = tensors[0].shallow_copy()
         result.share = torch_stack(
@@ -630,9 +630,9 @@ def scatter_(self, dim, index, src):
         """
         if is_tensor(src):
             src = ArithmeticSharedTensor(src)
-        assert isinstance(
-            src, ArithmeticSharedTensor
-        ), "Unrecognized scatter src type: %s" % type(src)
+        assert isinstance(src, ArithmeticSharedTensor), (
+            "Unrecognized scatter src type: %s" % type(src)
+        )
         self.share.scatter_(dim, index, src.share)
         return self
 

crypten/mpc/primitives/binary.py

@@ -318,9 +318,9 @@ def __setitem__(self, index, value):
         """Set tensor values by index"""
         if is_tensor(value) or isinstance(value, list):
             value = BinarySharedTensor(value)
-        assert isinstance(
-            value, BinarySharedTensor
-        ), "Unsupported input type %s for __setitem__" % type(value)
+        assert isinstance(value, BinarySharedTensor), (
+            "Unsupported input type %s for __setitem__" % type(value)
+        )
         self.share.__setitem__(index, value.share)
 
     @staticmethod
@@ -436,9 +436,9 @@ def scatter_(self, dim, index, src):
         """
         if is_tensor(src):
             src = BinarySharedTensor(src)
-        assert isinstance(
-            src, BinarySharedTensor
-        ), "Unrecognized scatter src type: %s" % type(src)
+        assert isinstance(src, BinarySharedTensor), (
+            "Unrecognized scatter src type: %s" % type(src)
+        )
         self.share.scatter_(dim, index, src.share)
         return self
 

crypten/mpc/primitives/converters.py

@@ -16,7 +16,6 @@
 
 
 def _A2B(arithmetic_tensor):
-
     # first try memory-inefficient implementation that takes O(log P) rounds:
     try:
         binary_tensor = BinarySharedTensor.stack(

crypten/mpc/provider/ttp_provider.py

@@ -309,7 +309,6 @@ def _get_binary_PRSS(self, size, bitlength=None, remove_rank=None):
         return result
 
     def additive(self, size0, size1, op, *args, **kwargs):
-
         # Add all shares of `a` and `b` to get plaintext `a` and `b`
         a = self._get_additive_PRSS(size0)
         b = self._get_additive_PRSS(size1)

crypten/nn/module.py

@@ -470,7 +470,6 @@ def _apply(self, fn):
     def encrypt(self, mode=True, src=0):
         """Encrypts the model."""
         if mode != self.encrypted:
-
             # encrypt / decrypt parameters:
             self.encrypted = mode
             for name, param in self.named_parameters(recurse=False):
@@ -705,7 +704,6 @@ def _clear_unused_values():
             _mark_as_computed(input_name)
         node_to_compute = _find_computable_node()
         while node_to_compute is not None:
-
             # compute output of module:
             input = [values[name] for name in self._graph[node_to_compute]]
             if len(input) == 1:
@@ -726,8 +724,8 @@ def _clear_unused_values():
                 assert isinstance(
                     output, tuple
                 ), f"expected outputs {output_names} of {module} to be tuple, not {type(output)}"
-                assert len(output_names) == len(
-                    output
+                assert (
+                    len(output_names) == len(output)
                 ), f"expected {len(output_names)} outputs from {module}, received {len(output)}"
                 for node, value in zip(output_names, output):
                     values[node] = value
@@ -1381,7 +1379,6 @@ class Expand(Module):
     """
 
     def forward(self, x):
-
         # unpack inputs:
         input, shape = tuple(x)
         if torch.is_tensor(shape):
@@ -1966,7 +1963,6 @@ def __init__(self, stride, padding, dilation, groups=1):
         self.groups = groups
 
     def forward(self, x):
-
         # unpack inputs:
         if len(x) == 2:
             x, weight = x
@@ -2010,7 +2006,6 @@ def forward(self, x):
 
     @staticmethod
     def from_onnx(attributes=None):
-
         # check attribute inputs:
         if attributes is None:
             attributes = {}
@@ -2124,7 +2119,6 @@ def __init__(
         groups=1,
         bias=True,
     ):
-
         # check inputs:
         super().__init__()
         assert isinstance(stride, int), "stride must be an integer"
@@ -2558,7 +2552,6 @@ def forward(self, x):
 
     @staticmethod
     def from_onnx(pool_type, attributes=None):
-
         # check attributes:
         if attributes is None:
             attributes = {}
@@ -2887,7 +2880,6 @@ def forward(self, input):
 
     @staticmethod
     def from_onnx(parameters=None, attributes=None):
-
         # preprocess all attributes:
         if parameters is None:
             parameters = {}

crypten/nn/onnx_converter.py

@@ -178,7 +178,6 @@ def _to_crypten(onnx_model):
 
     # loop over all nodes:
     for node in onnx_model.graph.node:
-
         # get attributes and node type:
         attributes = {attr.name: _get_attribute_value(attr) for attr in node.attribute}
         crypten_class = _get_operator_class(node.op_type, attributes)
@@ -271,21 +270,21 @@ def _update_onnx_symbolic_registry():
         for version_key, version_val in sym_registry._registry.items():
             for function_key in version_val.keys():
                 if function_key == "softmax":
-                    sym_registry._registry[version_key][
-                        function_key
-                    ] = _onnx_crypten_softmax
+                    sym_registry._registry[version_key][function_key] = (
+                        _onnx_crypten_softmax
+                    )
                 if function_key == "log_softmax":
-                    sym_registry._registry[version_key][
-                        function_key
-                    ] = _onnx_crypten_logsoftmax
+                    sym_registry._registry[version_key][function_key] = (
+                        _onnx_crypten_logsoftmax
+                    )
                 if function_key == "dropout":
-                    sym_registry._registry[version_key][
-                        function_key
-                    ] = _onnx_crypten_dropout
+                    sym_registry._registry[version_key][function_key] = (
+                        _onnx_crypten_dropout
+                    )
                 if function_key == "feature_dropout":
-                    sym_registry._registry[version_key][
-                        function_key
-                    ] = _onnx_crypten_feature_dropout
+                    sym_registry._registry[version_key][function_key] = (
+                        _onnx_crypten_feature_dropout
+                    )
     else:
         # Update ONNX symbolic registry using torch.onnx.register_custom_op_symbolic
         torch.onnx.register_custom_op_symbolic(

crypten/nn/tensorboard.py

@@ -50,7 +50,6 @@ def graph(model):
 
     # loop all graph connections:
     for output_name, input_names in model._graph.items():
-
         # get parameters and type of module:
         module = modules[output_name]
         op = str(type(module))

examples/bandits/launcher.py

@@ -380,7 +380,6 @@ def build_learner(args, bandits, download_mnist):
             logging.info("Loading clusters from file...")
             clusters = torch.load(clusters_file)
         else:
-
             # load data and allocate clusters:
             context, _ = load_data(
                 split=args.mnist_split, pca=pca, download_mnist_func=download_mnist

examples/bandits/plain_contextual_bandits.py

@@ -50,7 +50,6 @@ def online_learner(
 
         # initialization of model parameters:
         if idx == 0:
-
             # initialize accumulators for linear least squares:
             A_inv = torch.stack(
                 [

examples/bandits/private_contextual_bandits.py

@@ -62,7 +62,6 @@ def online_learner(
 
         # initialization of model parameters:
         if idx == 0:
-
             # initialize accumulators for linear least squares:
             A_inv = [torch.eye(num_features).unsqueeze(0) for _ in range(num_arms)]
             A_inv = crypten.cat([crypten.cryptensor(A) for A in A_inv])