Install dependencies for website build (#558)

Summary:
As reported in #557, API reference page on our website was broken (nothing to link because it's fixed now)

Here's an [example](https://github.com/pytorch/opacus/actions/runs/3992500542/jobs/6848405307) of a failed build:

```
WARNING: autodoc: failed to import module 'accountants.gdp' from module 'opacus'; the following exception was raised:
No module named 'numpy'
WARNING: autodoc: failed to import module 'accountants.accountant' from module 'opacus'; the following exception was raised:
No module named 'numpy'
WARNING: autodoc: failed to import module 'accountants.rdp' from module 'opacus'; the following exception was raised:
No module named 'numpy'
WARNING: autodoc: failed to import module 'accountants.utils' from module 'opacus'; the following exception was raised:
No module named 'numpy'
```

This PR updates dependencies for the website building action

Pull Request resolved: #558

Reviewed By: karthikprasad

Differential Revision: D42997411

Pulled By: ffuuugor

fbshipit-source-id: a2b925bf66fdca9e48c4271071615f2a1c0e0f84

.github/workflows/deploy.yml

@@ -17,7 +17,7 @@ jobs:
       - name: Install dependencies
         run: |
           python -m pip install --upgrade pip
-          python -m pip install sphinx sphinx-autodoc-typehints nbsphinx bs4
+          python -m pip install opacus sphinx sphinx-autodoc-typehints nbsphinx bs4
           yarn install
       - name: Build website
         run: ./scripts/build_website.sh -b

.github/workflows/test-deploy.yml

@@ -19,7 +19,7 @@ jobs:
       - name: Install dependencies
         run: |
           python -m pip install --upgrade pip
-          python -m pip install sphinx sphinx-autodoc-typehints nbsphinx bs4
+          python -m pip install opacus sphinx sphinx-autodoc-typehints nbsphinx bs4
           yarn install
       - name: Test build website
         run: ./scripts/build_website.sh -b

benchmarks/benchmark_layer.py

@@ -90,7 +90,6 @@ def run_layer_benchmark(
 
 
 def main(args) -> None:
-
     with open(args.config_file) as config_file:
         config = json.load(config_file)
 

benchmarks/run_benchmarks.py

@@ -83,7 +83,6 @@ def run_and_save_benchmark(
 
 
 def main(args) -> None:
-
     if args.verbose:
         logger.setLevel(logging.DEBUG)
 

benchmarks/tests/test_utils.py

@@ -150,7 +150,6 @@ def test_get_path(config: Dict[str, Any], path: str) -> None:
 def pickle_data_and_config(
     config: Dict[str, Any], root: str, suffix: str
 ) -> Tuple[Dict[str, Any], Dict[str, Any]]:
-
     # setup test directory and save results to pickle file
     os.mkdir(root)
     try:

examples/cifar10.py

@@ -159,7 +159,6 @@ def compute_loss_stateless_model(params, sample, target):
         params = list(model.parameters())
 
     for i, (images, target) in enumerate(tqdm(train_loader)):
-
         images = images.to(device)
         target = target.to(device)
 
@@ -172,7 +171,7 @@ def compute_loss_stateless_model(params, sample, target):
             )
             per_sample_grads = [g.detach() for g in per_sample_grads]
             loss = torch.mean(per_sample_losses)
-            for (p, g) in zip(params, per_sample_grads):
+            for p, g in zip(params, per_sample_grads):
                 p.grad_sample = g
         else:
             loss = criterion(output, target)

examples/vision_benchmark.py

@@ -134,7 +134,7 @@ def main():  # noqa: C901
         start = time.time()
         data_time = 0
         data_end = time.time()
-        for (images, target) in tqdm(train_loader):
+        for images, target in tqdm(train_loader):
             data_time += time.time() - data_end
             images = images.cuda(non_blocking=True)
 

opacus/optimizers/perlayeroptimizer.py

@@ -55,7 +55,7 @@ def __init__(
         )
 
     def clip_and_accumulate(self):
-        for (p, max_grad_norm) in zip(self.params, self.max_grad_norms):
+        for p, max_grad_norm in zip(self.params, self.max_grad_norms):
             _check_processed_flag(p.grad_sample)
 
             grad_sample = self._get_flat_grad_sample(p)

opacus/tests/accountants_test.py

@@ -138,7 +138,6 @@ def test_get_noise_multiplier_prv_steps(self):
     )
     @settings(deadline=10000)
     def test_get_noise_multiplier_overshoot(self, epsilon, epochs, sample_rate, delta):
-
         noise_multiplier = get_noise_multiplier(
             target_epsilon=epsilon,
             target_delta=delta,

opacus/tests/ddp_hook_check.py

@@ -298,7 +298,6 @@ def test_ddp_hook(self):
         )
 
     def test_add_remove_ddp_hooks(self):
-
         remaining_hooks = {
             "attached": None,
             "detached": None,

opacus/tests/dp_layers/common.py

@@ -352,7 +352,6 @@ def _check_packed_sequence(
         rtol: float,
         failure_msgs: Optional[Sequence] = None,
     ) -> bool:
-
         try:
             padded_seq_nn, seq_lens_nn = pad_packed_sequence(nn_out, batch_first_nn)
         except ValueError:

opacus/tests/grad_sample_module_test.py

@@ -53,7 +53,6 @@ def name(self):
 
 
 class GradSampleModuleTest(unittest.TestCase):
-
     CLS = GradSampleModule
 
     def setUp(self):

opacus/tests/grad_samples/common.py

@@ -249,7 +249,6 @@ def run_test(
 
         for grad_sample_mode in grad_sample_modes:
             for loss_reduction in ["sum", "mean"]:
-
                 with self.subTest(
                     grad_sample_mode=grad_sample_mode, loss_reduction=loss_reduction
                 ):

opacus/tests/grad_samples/conv1d_test.py

@@ -48,7 +48,6 @@ def test_conv1d(
         dilation: int,
         groups: int,
     ):
-
         if padding == "same" and stride != 1:
             return
         out_channels = out_channels_mapper(C)

opacus/tests/grad_samples/conv3d_test.py

@@ -52,7 +52,6 @@ def test_conv3d(
         dilation: int,
         groups: int,
     ):
-
         if padding == "same" and stride != 1:
             return
         out_channels = out_channels_mapper(C)

opacus/tests/grad_samples/dp_multihead_attention_test.py

@@ -66,7 +66,6 @@ def test_multihead_attention(
         add_zero_attn: bool,
         kv_dim: bool,
     ):
-
         if kv_dim:
             kdim, vdim = D, D
         else:

opacus/tests/grad_samples/embedding_test.py

@@ -44,7 +44,6 @@ def test_input_across_dims(
         dim: int,
         batch_first: bool,
     ):
-
         if dim == 1:  # TODO: fix when dim is 1
             size = [T]
         elif dim == 2:

opacus/tests/grad_samples/group_norm_test.py

@@ -45,7 +45,6 @@ def test_3d_input_groups(
         W: int,
         num_groups: Union[int, str],
     ):
-
         if num_groups == "C":
             num_groups = C
 

opacus/tests/grad_samples/instance_norm1d_test.py

@@ -34,7 +34,6 @@ def test_3d_input(
         C: int,
         W: int,
     ):
-
         x = torch.randn([N, C, W])
         norm = nn.InstanceNorm1d(num_features=C, affine=True, track_running_stats=False)
         self.run_test(x, norm, batch_first=True)

opacus/tests/grad_samples/instance_norm2d_test.py

@@ -36,7 +36,6 @@ def test_4d_input(
         W: int,
         H: int,
     ):
-
         x = torch.randn([N, C, H, W])
         norm = nn.InstanceNorm2d(num_features=C, affine=True, track_running_stats=False)
         self.run_test(x, norm, batch_first=True)

opacus/tests/grad_samples/layer_norm_test.py

@@ -40,7 +40,6 @@ def test_input_norm(
         input_dim: int,
         norm_dim: int,
     ):
-
         if norm_dim >= input_dim:
             return
         if norm_dim == 1:

opacus/tests/grad_samples/linear_test.py

@@ -42,7 +42,6 @@ def test_input_bias(
         bias: bool,
         batch_first: bool,
     ):
-
         if input_dim == 2:
             if not batch_first:
                 return  # see https://github.com/pytorch/opacus/pull/265

opacus/tests/grad_samples/sequence_bias_test.py

@@ -36,7 +36,6 @@ def test_batch_second(
         D: int,
         batch_first: bool,
     ):
-
         seqbias = SequenceBias(D, batch_first)
         if batch_first:
             x = torch.randn([N, T, D])

opacus/tests/privacy_engine_test.py

@@ -150,7 +150,6 @@ def _train_steps(
         dl: DataLoader,
         max_steps: Optional[int] = None,
     ):
-
         steps = 0
         epochs = 1 if max_steps is None else math.ceil(max_steps / len(dl))