Merge branch 'main' into improve_ci_speed

* main:
  Use better naming for CI jobs (#2110)
  Update Opacus example to use latest version (#2112)
  Fix quickstart-fastai example (#2111)

.github/workflows/flower_tool.yml → .github/workflows/devtools.yml

@@ -1,4 +1,4 @@
-name: Build Tools
+name: DevTools
 
 on:
   push:
@@ -18,6 +18,7 @@ env:
 jobs:
   test_tools:
     runs-on: ubuntu-22.04
+    name: Test
     steps:
       - uses: actions/checkout@v3
       - name: Bootstrap

.github/workflows/docs.yml

@@ -14,6 +14,7 @@ env:
 jobs:
   build_docs:
     runs-on: ubuntu-22.04
+    name: Build
     steps:
       - uses: actions/checkout@v3
       - name: Bootstrap
@@ -33,6 +34,7 @@ jobs:
     needs: build_docs
     if: github.ref == 'refs/heads/main'
     runs-on: ubuntu-22.04
+    name: Deploy 
     steps:
       - uses: actions/checkout@v3
       - name: Bootstrap

.github/workflows/flower.yml → .github/workflows/framework.yml

@@ -1,4 +1,4 @@
-name: Core
+name: Framework
 
 on:
   push:

.github/workflows/flower_nightly.yml → .github/workflows/release.yml

@@ -10,6 +10,7 @@ env:
 jobs:
   nightly_release:
     runs-on: ubuntu-22.04
+    name: Nightly
     steps:
       - uses: actions/checkout@v3
       - name: Bootstrap

.github/workflows/swift.yml

@@ -14,7 +14,7 @@ jobs:
     defaults:
         run:
             working-directory: src/swift/flwr
-    name: test_swift
+    name: Test
     strategy:
       matrix:
         os: [macos-latest]

e2e/opacus/pyproject.toml

@@ -11,6 +11,6 @@ authors = ["The Flower Authors <hello@flower.dev>"]
 [tool.poetry.dependencies]
 python = "^3.8"
 flwr = { path = "../../", develop = true, extras = ["simulation"] }
-opacus = "1.4.0"
+opacus = "^1.4.0"
 torch = "^1.13.1"
 torchvision = "^0.14.0"

examples/opacus/dp_cifar_client.py

@@ -29,5 +29,5 @@ def load_data():
 model = Net()
 trainloader, testloader, sample_rate = load_data()
 fl.client.start_numpy_client(
-    "127.0.0.1:8080", client=DPCifarClient(model, trainloader, testloader, sample_rate)
+    server_address="127.0.0.1:8080", client=DPCifarClient(model, trainloader, testloader)
 )

examples/opacus/dp_cifar_main.py

@@ -46,19 +46,17 @@ def forward(self, x: torch.Tensor) -> torch.Tensor:
         return x
 
 
-def train(net, trainloader, privacy_engine, epochs):
+def train(net, trainloader, privacy_engine, optimizer, epochs):
     criterion = torch.nn.CrossEntropyLoss()
-    optimizer = torch.optim.SGD(net.parameters(), lr=0.001, momentum=0.9)
-    privacy_engine.attach(optimizer)
     for _ in range(epochs):
         for images, labels in trainloader:
             images, labels = images.to(DEVICE), labels.to(DEVICE)
             optimizer.zero_grad()
             loss = criterion(net(images), labels)
             loss.backward()
             optimizer.step()
-    epsilon, _ = optimizer.privacy_engine.get_privacy_spent(
-        PRIVACY_PARAMS["target_delta"]
+    epsilon = privacy_engine.get_epsilon(
+        delta=PRIVACY_PARAMS["target_delta"]
     )
     return epsilon
 
@@ -79,16 +77,16 @@ def test(net, testloader):
 
 # Define Flower client.
 class DPCifarClient(fl.client.NumPyClient):
-    def __init__(self, model, trainloader, testloader, sample_rate) -> None:
+    def __init__(self, model, trainloader, testloader) -> None:
         super().__init__()
-        self.model = model
-        self.trainloader = trainloader
+        optimizer = torch.optim.SGD(model.parameters(), lr=0.001, momentum=0.9)
         self.testloader = testloader
         # Create a privacy engine which will add DP and keep track of the privacy budget.
-        self.privacy_engine = PrivacyEngine(
-            self.model,
-            sample_rate=sample_rate,
-            target_delta=PRIVACY_PARAMS["target_delta"],
+        self.privacy_engine = PrivacyEngine()
+        self.model, self.optimizer, self.trainloader = self.privacy_engine.make_private(
+            module=model,
+            optimizer=optimizer,
+            data_loader=trainloader,
             max_grad_norm=PRIVACY_PARAMS["max_grad_norm"],
             noise_multiplier=PRIVACY_PARAMS["noise_multiplier"],
         )
@@ -104,7 +102,7 @@ def set_parameters(self, parameters):
     def fit(self, parameters, config):
         self.set_parameters(parameters)
         epsilon = train(
-            self.model, self.trainloader, self.privacy_engine, PARAMS["local_epochs"]
+            self.model, self.trainloader, self.privacy_engine, self.optimizer, PARAMS["local_epochs"]
         )
         print(f"epsilon = {epsilon:.2f}")
         return (

examples/opacus/dp_cifar_simulation.py

@@ -45,9 +45,7 @@ def client_fn(cid: str) -> fl.client.Client:
     client_trainloader = DataLoader(client_trainset, PARAMS["batch_size"])
     client_testloader = DataLoader(client_testset, PARAMS["batch_size"])
 
-    sample_rate = PARAMS["batch_size"] / len(client_trainset)
-
-    return DPCifarClient(model, client_trainloader, client_testloader, sample_rate)
+    return DPCifarClient(model, client_trainloader, client_testloader)
 
 
 # Define an evaluation function for centralized evaluation (using whole CIFAR10 testset).

examples/opacus/pyproject.toml

@@ -10,8 +10,8 @@ authors = ["The Flower Authors <hello@flower.dev>"]
 
 [tool.poetry.dependencies]
 python = "^3.8"
-flwr = "^0.17.0"
+flwr = "^1.0.0"
 # flwr = { path = "../../", develop = true }  # Development
-opacus = "~0.14.0"
+opacus = "^1.4.0"
 torch = "^1.13.1"
-torchvision = "~0.8.0"
+torchvision = "^0.14.0"

examples/quickstart_fastai/client.py

@@ -14,7 +14,7 @@
 
 # Load dataset
 dls = ImageDataLoaders.from_folder(
-    path, valid_pct=0.5, train="training", valid="testing"
+    path, valid_pct=0.5, train="training", valid="testing", num_workers=0
 )
 
 # Define model
@@ -24,7 +24,7 @@
 # Define Flower client
 class FlowerClient(fl.client.NumPyClient):
     def get_parameters(self, config):
-        return [val.numpy() for _, val in learn.model.state_dict().items()]
+        return [val.cpu().numpy() for _, val in learn.model.state_dict().items()]
 
     def set_parameters(self, parameters):
         params_dict = zip(learn.model.state_dict().keys(), parameters)