Merge branch 'main' into fsi_update

Author: holgerroth

docs/user_guide/data_scientist_guide/flower_integration/flower_job_structure.rst

@@ -43,7 +43,7 @@ Here is an example of ``pyproject.toml``, taken from :github_nvflare_link:`this
     description = ""
     license = "Apache-2.0"
     dependencies = [
-        "flwr>=1.16,<1.26",
+        "flwr>=1.26",
         "nvflare~=2.5.0rc",
         "torch==2.2.1",
         "torchvision==0.17.1",
@@ -62,15 +62,14 @@ Here is an example of ``pyproject.toml``, taken from :github_nvflare_link:`this
     [tool.flwr.app.config]
     num-server-rounds = 3
 
-    [tool.flwr.federations]
-    default = "local-simulation"
-
-    [tool.flwr.federations.local-simulation]
-    options.num-supernodes = 2
-
 
 .. note:: Note that the information defined in pyproject.toml must match the code in the project folder!
 
+.. note:: For NVFlare-managed Flower jobs, NVFlare creates a job-scoped
+   ``$FLWR_HOME/config.toml`` with the SuperLink connection details. You do not
+   need to define a ``[tool.flwr.federations]`` section in ``pyproject.toml`` for
+   FLARE execution.
+
 Project Name
 ~~~~~~~~~~~~
 The project name should match the name of the project folder, though not a requirement. In this example, it is ``flwr_pt``. 
@@ -136,13 +135,9 @@ The content of this section is specific to the server app code. The ``server.py`
 Note that you can also pass `run_config` arguments directly through the job definition via 
 `FlowerRecipe(..., run_config={"num-server-rounds": 5})` to override the default values listed in `pyproject.toml`.
 
-Supernode Count
-~~~~~~~~~~~~~~~
-If you run the Flower job with its simulation (not as a FLARE job), you need to specify how many clients (supernodes) to use
-for the simulation in the ``[tool.flwr.federations.local-simulation]`` section, like this:
-
-.. code-block:: toml
-
-    options.num-supernodes = 2
-
-But this does not apply when submitting it as a FLARE job.
+Simulation Profiles
+~~~~~~~~~~~~~~~~~~~
+If you run the Flower job directly with Flower simulation instead of submitting it
+as a FLARE job, configure the simulation profile using Flower's current simulation
+configuration mechanism. This is separate from the NVFlare execution path, where
+NVFlare manages the SuperLink connection through ``$FLWR_HOME/config.toml``.

docs/user_guide/data_scientist_guide/flower_integration/flower_run_as_flare_job.rst

@@ -4,11 +4,17 @@ Run Flower Application as FLARE Job
 
 Before running Flower applications with FLARE, you must have both FLARE and Flower frameworks
 installed in your Python environment. NVFlare 2.7 Flower integration currently requires
-``flwr>=1.16,<1.26``.
+``flwr>=1.26``.
 
 .. code-block:: shell
 
-    pip install 'flwr>=1.16,<1.26'
+    pip install 'flwr>=1.26'
+
+With Flower 1.26 and newer, SuperLink connections are no longer configured in
+``pyproject.toml`` for NVFlare jobs. Instead, NVFlare writes a job-scoped Flower
+configuration file at ``$FLWR_HOME/config.toml`` and uses it to connect
+``flwr run``, ``flwr list``, and ``flwr stop`` to the job's dynamically assigned
+SuperLink control API address.
 
 To run a Flower application as a job in FLARE, follow these steps:
 

examples/hello-world/hello-flower/flwr-pt-tb/flwr_pt_tb/task.py

@@ -11,9 +11,11 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import warnings
 from collections import OrderedDict
 from logging import INFO
 
+import numpy as np
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
@@ -24,6 +26,13 @@
 
 DEVICE = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
+try:
+    VISIBLE_DEPRECATION_WARNING = np.exceptions.VisibleDeprecationWarning
+except AttributeError:
+    VISIBLE_DEPRECATION_WARNING = Warning
+
+CIFAR10_NUMPY_PICKLE_WARNING = r".*align should be passed as Python or NumPy boolean.*align=0.*"
+
 
 class Net(nn.Module):
     """Model (simple CNN adapted from 'PyTorch: A 60 Minute Blitz')"""
@@ -49,11 +58,23 @@ def forward(self, x: torch.Tensor) -> torch.Tensor:
 def load_data():
     """Load CIFAR-10 (training and test set)."""
     trf = Compose([ToTensor(), Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
-    trainset = CIFAR10("./data", train=True, download=True, transform=trf)
-    testset = CIFAR10("./data", train=False, download=True, transform=trf)
+    trainset = _load_cifar10_dataset(train=True, transform=trf)
+    testset = _load_cifar10_dataset(train=False, transform=trf)
     return DataLoader(trainset, batch_size=32, shuffle=True), DataLoader(testset)
 
 
+def _load_cifar10_dataset(train: bool, transform):
+    # NumPy 2.4 warns when unpickling CIFAR-10's legacy dtype metadata, but
+    # torchvision still reads the dataset correctly.
+    with warnings.catch_warnings():
+        warnings.filterwarnings(
+            "ignore",
+            message=CIFAR10_NUMPY_PICKLE_WARNING,
+            category=VISIBLE_DEPRECATION_WARNING,
+        )
+        return CIFAR10("./data", train=train, download=True, transform=transform)
+
+
 def train(net, trainloader, valloader, epochs, device, learning_rate, momentum):
     """Train the model on the training set."""
     log(INFO, "Starting training...")

examples/hello-world/hello-flower/flwr-pt-tb/pyproject.toml

@@ -3,10 +3,10 @@ requires = ["hatchling"]
 build-backend = "hatchling.build"
 
 # Tested with:
-#   flwr==1.20.0
-#   nvflare==2.6.1
-#   torch==2.7.1
-#   torchvision==0.22.1
+#   flwr==1.27.0
+#   nvflare==2.7.2
+#   torch==2.11.0
+#   torchvision==0.26.0
 #   tensorboard==2.20.0
 
 [project]
@@ -15,7 +15,7 @@ version = "1.0.0"
 description = ""
 license = "Apache-2.0"
 dependencies = [
-    "flwr>=1.16,<1.26",
+    "flwr>=1.26",
     "nvflare~=2.6.0rc",
     "torch",
     "torchvision",
@@ -36,11 +36,3 @@ clientapp = "flwr_pt_tb.client:app"
 num-server-rounds = 3
 learning-rate = 0.001
 momentum = 0.9
-
-[tool.flwr.federations]
-default = "local-simulation"
-
-[tool.flwr.federations.local-simulation]
-options.num-supernodes = 2
-address = "127.0.0.1:9093"
-insecure = true

examples/hello-world/hello-flower/flwr-pt/flwr_pt/task.py

@@ -11,9 +11,11 @@
 # WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
 # See the License for the specific language governing permissions and
 # limitations under the License.
+import warnings
 from collections import OrderedDict
 from logging import INFO
 
+import numpy as np
 import torch
 import torch.nn as nn
 import torch.nn.functional as F
@@ -24,6 +26,13 @@
 
 DEVICE = torch.device("cuda:0" if torch.cuda.is_available() else "cpu")
 
+try:
+    VISIBLE_DEPRECATION_WARNING = np.exceptions.VisibleDeprecationWarning
+except AttributeError:
+    VISIBLE_DEPRECATION_WARNING = Warning
+
+CIFAR10_NUMPY_PICKLE_WARNING = r".*align should be passed as Python or NumPy boolean.*align=0.*"
+
 
 class Net(nn.Module):
     """Model (simple CNN adapted from 'PyTorch: A 60 Minute Blitz')"""
@@ -49,11 +58,23 @@ def forward(self, x: torch.Tensor) -> torch.Tensor:
 def load_data():
     """Load CIFAR-10 (training and test set)."""
     trf = Compose([ToTensor(), Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])
-    trainset = CIFAR10("./data", train=True, download=True, transform=trf)
-    testset = CIFAR10("./data", train=False, download=True, transform=trf)
+    trainset = _load_cifar10_dataset(train=True, transform=trf)
+    testset = _load_cifar10_dataset(train=False, transform=trf)
     return DataLoader(trainset, batch_size=32, shuffle=True), DataLoader(testset)
 
 
+def _load_cifar10_dataset(train: bool, transform):
+    # NumPy 2.4 warns when unpickling CIFAR-10's legacy dtype metadata, but
+    # torchvision still reads the dataset correctly.
+    with warnings.catch_warnings():
+        warnings.filterwarnings(
+            "ignore",
+            message=CIFAR10_NUMPY_PICKLE_WARNING,
+            category=VISIBLE_DEPRECATION_WARNING,
+        )
+        return CIFAR10("./data", train=train, download=True, transform=transform)
+
+
 def train(net, trainloader, valloader, epochs, device, learning_rate, momentum):
     """Train the model on the training set."""
     log(INFO, "Starting training...")

examples/hello-world/hello-flower/flwr-pt/pyproject.toml

@@ -3,10 +3,10 @@ requires = ["hatchling"]
 build-backend = "hatchling.build"
 
 # Tested with:
-#   flwr==1.20.0
-#   nvflare==2.6.1
-#   torch==2.7.1
-#   torchvision==0.22.1
+#   flwr==1.27.0
+#   nvflare==2.7.2
+#   torch==2.11.0
+#   torchvision==0.26.0
 #   tensorboard==2.20.0
 
 [project]
@@ -15,7 +15,7 @@ version = "1.0.0"
 description = ""
 license = "Apache-2.0"
 dependencies = [
-    "flwr>=1.16,<1.26",
+    "flwr>=1.26",
     "nvflare~=2.6.0rc",
     "torch",
     "torchvision",
@@ -36,11 +36,3 @@ clientapp = "flwr_pt.client:app"
 num-server-rounds = 3
 learning-rate = 0.001
 momentum = 0.9
-
-[tool.flwr.federations]
-default = "local-simulation"
-
-[tool.flwr.federations.local-simulation]
-options.num-supernodes = 2
-address = "127.0.0.1:9093"
-insecure = true

examples/hello-world/hello-flower/requirements.txt

@@ -1,5 +1,5 @@
 nvflare~=2.7.2rc
-flwr>=1.16,<1.26
+flwr>=1.26
 torch
 torchvision
 tensorboard