From cf3b5ce74748220aaf87928d2949cddde6a71d29 Mon Sep 17 00:00:00 2001
From: nvkevlu <55759229+nvkevlu@users.noreply.github.com>
Date: Mon, 14 Oct 2024 18:01:44 +0000
Subject: [PATCH] =?UTF-8?q?Deploying=20to=20gh-pages=20from=20@=20NVIDIA/N?=
 =?UTF-8?q?VFlare@68c7eb8a2df35afce27ec8c0783446ba05c977e2=20=F0=9F=9A=80?=
MIME-Version: 1.0
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit

---
 version/2.5/index.html | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/version/2.5/index.html b/version/2.5/index.html
index 097d1c742f..ab79a2e424 100644
--- a/version/2.5/index.html
+++ b/version/2.5/index.html
@@ -27,7 +27,7 @@
 Server
 </span> </button> </li> <li class="flex-1" role="presentation"> <button class="w-36 md:w-36 inline-block p-5 border-r-0 md:border-r-2 md:border-b-0 font-bold text-left" id="job-tab" data-tabs-target="#job-wrapper" type="button" role="tab" aria-controls="job" aria-selected="false"> <span id="default-message" class="inline-flex items-center py-2.5"> <svg class="w-8 h-8 text-gray-800 dark:text-white mr-2" aria-hidden="true" xmlns="http://www.w3.org/2000/svg" width="24" height="24" fill="none" viewBox="0 0 24 24"> <path stroke="" stroke-linecap="round" stroke-linejoin="round" stroke-width="2" d="M10 3v4a1 1 0 0 1-1 1H5m5 4-2 2 2 2m4-4 2 2-2 2m5-12v16a1 1 0 0 1-1 1H6a1 1 0 0 1-1-1V7.914a1 1 0 0 1 .293-.707l3.914-3.914A1 1 0 0 1 9.914 3H18a1 1 0 0 1 1 1Z"></path> </svg>
 Job
-</span> </button> </li> </ul> <!-- Client wrapper --> <div id="client-wrapper" class="py-4 text-left overflow-x-auto"></div> <!-- Server wrapper --> <div id="server-wrapper" class="py-4 text-left overflow-x-auto hidden"></div> <!-- Job wrapper --> <div id="job-wrapper" class="py-4 text-left overflow-x-auto hidden"></div> </div> <!-- Run wrapper --> <div id="run-wrapper" class="mx-auto max-w-5xl py-4 text-left"></div> </div> </div>     <script type="module">const frameworks = [{"id":"pytorch","colab_link":"https://colab.research.google.com/github/NVIDIA/NVFlare/blob/2.5/examples/getting_started/pt/nvflare_pt_getting_started.ipynb","github_link":"https://github.com/NVIDIA/NVFlare/blob/2.5/examples/getting_started/pt/nvflare_pt_getting_started.ipynb","sections":[{"id":"install-pytorch","type":"install","framework":"pytorch","title":"Installation","description":"Install the required PyTorch dependencies for this example.","code":"\npip install nvflare torch torchvision\n"},{"id":"client-pytorch","type":"client","framework":"pytorch","title":"Client Code (cifar10_pt_fl.py)","description":"We use the Client API to convert the centralized training PyTorch code into federated learning code with only a few lines of changes highlighted below. Essentially the client will receive a model from NVIDIA FLARE, perform local training and validation, and then send the model back.","code":"\nimport torch\nimport torch.nn as nn\nimport torch.optim as optim\nimport torchvision\nimport torchvision.transforms as transforms\n\n\nclass Net(nn.Module):\n    def __init__(self):\n        super().__init__()\n        self.conv1 = nn.Conv2d(3, 6, 5)\n        self.pool = nn.MaxPool2d(2, 2)\n        self.conv2 = nn.Conv2d(6, 16, 5)\n        self.fc1 = nn.Linear(16 * 5 * 5, 120)\n        self.fc2 = nn.Linear(120, 84)\n        self.fc3 = nn.Linear(84, 10)\n\n    def forward(self, x):\n        x = self.pool(F.relu(self.conv1(x)))\n        x = self.pool(F.relu(self.conv2(x)))\n        x = torch.flatten(x, 1) # flatten all dimensions except batch\n        x = F.relu(self.fc1(x))\n        x = F.relu(self.fc2(x))\n        x = self.fc3(x)\n        return x\n\n\n# (1) import nvflare client API\nimport nvflare.client as flare\n\n# (optional) metrics\nfrom nvflare.client.tracking import SummaryWriter\n\n# (optional) set a fix place so we don't need to download everytime\nDATASET_PATH = \"/tmp/nvflare/data\"\n# If available, we use GPU to speed things up.\nDEVICE = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n\n\ndef main():\n    transform = transforms.Compose([\n      transforms.ToTensor(),\n      transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))\n    ])\n\n    batch_size = 4\n    epochs = 2\n\n    trainset = torchvision.datasets.CIFAR10(root=DATASET_PATH, train=True, download=True, transform=transform)\n    trainloader = torch.utils.data.DataLoader(trainset, batch_size=batch_size, shuffle=True, num_workers=2)\n\n    testset = torchvision.datasets.CIFAR10(root=DATASET_PATH, train=False, download=True, transform=transform)\n    testloader = torch.utils.data.DataLoader(testset, batch_size=batch_size, shuffle=False, num_workers=2)\n\n    net = Net()\n\n    # (2) initializes NVFlare client API\n    flare.init()\n\n    summary_writer = SummaryWriter()\n    while flare.is_running():\n        # (3) receives FLModel from NVFlare\n        input_model = flare.receive()\n        print(f\"current_round={input_model.current_round}\")\n\n        # (4) loads model from NVFlare\n        net.load_state_dict(input_model.params)\n\n        criterion = nn.CrossEntropyLoss()\n        optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)\n\n        # (optional) use GPU to speed things up\n        net.to(DEVICE)\n        # (optional) calculate total steps\n        steps = epochs * len(trainloader)\n        for epoch in range(epochs):  # loop over the dataset multiple times\n\n            running_loss = 0.0\n            for i, data in enumerate(trainloader, 0):\n                # get the inputs; data is a list of [inputs, labels]\n                # (optional) use GPU to speed things up\n                inputs, labels = data[0].to(DEVICE), data[1].to(DEVICE)\n\n                # zero the parameter gradients\n                optimizer.zero_grad()\n\n                # forward + backward + optimize\n                outputs = net(inputs)\n                loss = criterion(outputs, labels)\n                loss.backward()\n                optimizer.step()\n\n                # print statistics\n                running_loss += loss.item()\n                if i % 2000 == 1999:  # print every 2000 mini-batches\n                    print(f\"[{epoch + 1}, {i + 1:5d}] loss: {running_loss / 2000:.3f}\")\n                    global_step = input_model.current_round * steps + epoch * len(trainloader) + i\n\n                    summary_writer.add_scalar(\n                        tag=\"loss_for_each_batch\",\n                        scalar=running_loss, \n                        global_step=global_step\n                    )\n                    running_loss = 0.0\n\n        print(\"Finished Training\")\n\n        PATH = \"./cifar_net.pth\"\n        torch.save(net.state_dict(), PATH)\n\n        # (5) wraps evaluation logic into a method to re-use for\n        #       evaluation on both trained and received model\n        def evaluate(input_weights):\n            net = Net()\n            net.load_state_dict(input_weights)\n            # (optional) use GPU to speed things up\n            net.to(DEVICE)\n\n            correct = 0\n            total = 0\n            # since we're not training, we don't need to calculate the gradients for our outputs\n            with torch.no_grad():\n                for data in testloader:\n                    # (optional) use GPU to speed things up\n                    images, labels = data[0].to(DEVICE), data[1].to(DEVICE)\n                    # calculate outputs by running images through the network\n                    outputs = net(images)\n                    # the class with the highest energy is what we choose as prediction\n                    _, predicted = torch.max(outputs.data, 1)\n                    total += labels.size(0)\n                    correct += (predicted == labels).sum().item()\n\n            print(f\"Accuracy of the network on the 10000 test images: {100 * correct // total} %\")\n            return 100 * correct // total\n\n        # (6) evaluate on received model for model selection\n        accuracy = evaluate(input_model.params)\n        # (7) construct trained FL model\n        output_model = flare.FLModel(\n            params=net.cpu().state_dict(),\n            metrics={\"accuracy\": accuracy},\n            meta={\"NUM_STEPS_CURRENT_ROUND\": steps},\n        )\n        # (8) send model back to NVFlare\n        flare.send(output_model)\n\n\nif __name__ == \"__main__\":\n    main()","highlighted_lines":"29,58,61,63,139-143,145"},{"id":"server-pytorch","type":"server","framework":"pytorch","title":"Server Code (fedavg.py)","description":"The ModelController API is used to write a federated routine with mechanisms for distributing and receiving models from clients. Here we implement the basic FedAvg algorithm using some helper functions from BaseFedAvg.","code":"\nfrom nvflare.app_common.workflows.base_fedavg import BaseFedAvg\n\nclass FedAvg(BaseFedAvg):\n    def run(self) -> None:\n        self.info(\"Start FedAvg.\")\n\n        model = self.load_model()\n        model.start_round = self.start_round\n        model.total_rounds = self.num_rounds\n\n        for self.current_round in range(self.start_round, self.start_round + self.num_rounds):\n            self.info(f\"Round {self.current_round} started.\")\n            model.current_round = self.current_round\n\n            clients = self.sample_clients(self.num_clients)\n\n            results = self.send_model_and_wait(targets=clients, data=model)\n\n            aggregate_results = self.aggregate(results)\n\n            model = self.update_model(model, aggregate_results)\n\n            self.save_model(model)\n\n        self.info(\"Finished FedAvg.\")\n","highlighted_lines":"7,17,23"},{"id":"job-pytorch","type":"job","framework":"pytorch","title":"Job Code (fedavg_cifar10_pt_job.py)","description":"Lastly we construct the job with our 'cifar10_pt_fl.py' client script and 'FedAvg' server controller. The BaseFedJob automatically configures components for model persistence, model selection, and TensorBoard streaming. We then run the job with the FL simulator.","code":"\nfrom cifar10_pt_fl import Net\n\nfrom nvflare.app_common.workflows.fedavg import FedAvg\nfrom nvflare.app_opt.pt.job_config.base_fed_job import BaseFedJob\nfrom nvflare.job_config.script_runner import ScriptRunner\n\nif __name__ == \"__main__\":\n    n_clients = 2\n    num_rounds = 2\n    train_script = \"cifar10_pt_fl.py\"\n\n    # Create BaseFedJob with initial model\n    job = BaseFedJob(\n      name=\"cifar10_pt_fedavg\",\n      initial_model=Net(),\n    )\n\n    # Define the controller and send to server\n    controller = FedAvg(\n        num_clients=n_clients,\n        num_rounds=num_rounds,\n    )\n    job.to_server(controller)\n\n    # Add clients\n    for i in range(n_clients):\n        runner = ScriptRunner(script=train_script)\n        job.to(runner, f\"site-{i}\")\n\n    # job.export_job(\"/tmp/nvflare/jobs/job_config\")\n    job.simulator_run(\"/tmp/nvflare/jobs/workdir\", gpu=\"0\")\n"},{"id":"run-pytorch","type":"run","framework":"pytorch","title":"Run the Job","description":"To run the job with the simulator, copy the code and execute the job script, or run in Google Colab. Alternatively, export the job to a configuration and run the job in other modes.","code":"\npython3 fedavg_cifar10_pt_job.py\n"}]},{"id":"lightning","colab_link":"https://colab.research.google.com/github/NVIDIA/NVFlare/blob/2.5/examples/getting_started/pt/nvflare_lightning_getting_started.ipynb","github_link":"https://github.com/NVIDIA/NVFlare/blob/2.5/examples/getting_started/pt/nvflare_lightning_getting_started.ipynb","sections":[{"id":"install-lightning","type":"install","framework":"lightning","title":"Installation","description":"Install the required PyTorch Lightning dependencies for this example.","code":"\npip install nvflare pytorch_lightning\n"},{"id":"client-lightning","type":"client","framework":"lightning","title":"Client Code (cifar10_lightning_fl.py)","description":"We use the Client API to convert the centralized training PyTorch Lightning code into federated learning code with only a few lines of changes highlighted below. Essentially the client will receive a model from NVIDIA FLARE, perform local training and validation, and then send the model back.","code":"\nimport torch\nimport torch.nn.functional as F\nimport torch.optim as optim\nimport torchvision\nimport torchvision.transforms as transforms\nfrom pytorch_lightning import LightningDataModule, LightningModule Trainer, seed_everything\nfrom torch.utils.data import DataLoader, random_split\nfrom torchmetrics import Accuracy\n\n\nclass LitNet(LightningModule):\n    def __init__(self):\n        super().__init__()\n        self.save_hyperparameters()\n        self.model = Net()\n        self.train_acc = Accuracy(task=\"multiclass\", num_classes=NUM_CLASSES)\n        self.valid_acc = Accuracy(task=\"multiclass\", num_classes=NUM_CLASSES)\n        # (optional) pass additional information via self.__fl_meta__\n        self.__fl_meta__ = {}\n\n    def forward(self, x):\n        out = self.model(x)\n        return out\n\n    def training_step(self, batch, batch_idx):\n        x, labels = batch\n        outputs = self(x)\n        loss = criterion(outputs, labels)\n        self.train_acc(outputs, labels)\n        self.log(\"train_loss\", loss)\n        self.log(\"train_acc\", self.train_acc, on_step=True, on_epoch=False)\n        return loss\n\n    def evaluate(self, batch, stage=None):\n        x, labels = batch\n        outputs = self(x)\n        loss = criterion(outputs, labels)\n        self.valid_acc(outputs, labels)\n\n        if stage:\n            self.log(f\"{stage}_loss\", loss)\n            self.log(f\"{stage}_acc\", self.valid_acc, on_step=True, on_epoch=True)\n        return outputs\n\n    def validation_step(self, batch, batch_idx):\n        self.evaluate(batch, \"val\")\n\n    def test_step(self, batch, batch_idx):\n        self.evaluate(batch, \"test\")\n\n    def predict_step(self, batch: Any, batch_idx: int, dataloader_idx: int = 0) -> Any:\n        return self.evaluate(batch)\n\n    def configure_optimizers(self):\n        optimizer = optim.SGD(self.parameters(), lr=0.001, momentum=0.9)\n        return {\"optimizer\": optimizer}\n\n\nclass CIFAR10DataModule(LightningDataModule):\n    def __init__(self, data_dir: str = DATASET_PATH, batch_size: int = BATCH_SIZE):\n        super().__init__()\n        self.data_dir = data_dir\n        self.batch_size = batch_size\n\n    def prepare_data(self):\n        torchvision.datasets.CIFAR10(root=self.data_dir, train=True, download=True, transform=transform)\n        torchvision.datasets.CIFAR10(root=self.data_dir, train=False, download=True, transform=transform)\n\n    def setup(self, stage: str):\n        # Assign train/val datasets for use in dataloaders\n        if stage == \"fit\" or stage == \"validate\":\n            cifar_full = torchvision.datasets.CIFAR10(\n                root=self.data_dir, train=True, download=False, transform=transform\n            )\n            self.cifar_train, self.cifar_val = random_split(cifar_full, [0.8, 0.2])\n\n        # Assign test dataset for use in dataloader(s)\n        if stage == \"test\" or stage == \"predict\":\n            self.cifar_test = torchvision.datasets.CIFAR10(\n                root=self.data_dir, train=False, download=False, transform=transform\n            )\n\n    def train_dataloader(self):\n        return DataLoader(self.cifar_train, batch_size=self.batch_size)\n\n    def val_dataloader(self):\n        return DataLoader(self.cifar_val, batch_size=self.batch_size)\n\n    def test_dataloader(self):\n        return DataLoader(self.cifar_test, batch_size=self.batch_size)\n\n    def predict_dataloader(self):\n        return DataLoader(self.cifar_test, batch_size=self.batch_size)\n\n\n# (1) import nvflare lightning client API\nimport nvflare.client.lightning as flare\n\nseed_everything(7)\n\n\nDATASET_PATH = \"/tmp/nvflare/data\"\nBATCH_SIZE = 4\n\ntransform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])\n\ndef main():\n    model = LitNet()\n    cifar10_dm = CIFAR10DataModule()\n    if torch.cuda.is_available():\n        trainer = Trainer(max_epochs=1, accelerator=\"gpu\", devices=1 if torch.cuda.is_available() else None)\n    else:\n        trainer = Trainer(max_epochs=1, devices=None)\n\n    # (2) patch the lightning trainer\n    flare.patch(trainer)\n\n    while flare.is_running():\n        # (3) receives FLModel from NVFlare\n        # Note that we don't need to pass this input_model to trainer\n        # because after flare.patch the trainer.fit/validate will get the\n        # global model internally\n        input_model = flare.receive()\n        print(f\"\\n[Current Round={input_model.current_round}, Site = {flare.get_site_name()}]\\n\")\n\n        # (4) evaluate the current global model to allow server-side model selection\n        print(\"--- validate global model ---\")\n        trainer.validate(model, datamodule=cifar10_dm)\n\n        # perform local training starting with the received global model\n        print(\"--- train new model ---\")\n        trainer.fit(model, datamodule=cifar10_dm)\n\n        # test local model\n        print(\"--- test new model ---\")\n        trainer.test(ckpt_path=\"best\", datamodule=cifar10_dm)\n\n        # get predictions\n        print(\"--- prediction with new best model ---\")\n        trainer.predict(ckpt_path=\"best\", datamodule=cifar10_dm)\n\n\nif __name__ == \"__main__\":\n    main()\n","highlighted_lines":"97,116,118,123"},{"id":"server-lightning","type":"server","framework":"lightning","title":"Server Code (fedavg.py)","description":"The ModelController API is used to write a federated routine with mechanisms for distributing and receiving models from clients. Here we implement the basic FedAvg algorithm using some helper functions from BaseFedAvg..","code":"\nfrom nvflare.app_common.workflows.base_fedavg import BaseFedAvg\n\nclass FedAvg(BaseFedAvg):\n    def run(self) -> None:\n        self.info(\"Start FedAvg.\")\n\n        model = self.load_model()\n        model.start_round = self.start_round\n        model.total_rounds = self.num_rounds\n\n        for self.current_round in range(self.start_round, self.start_round + self.num_rounds):\n            self.info(f\"Round {self.current_round} started.\")\n            model.current_round = self.current_round\n\n            clients = self.sample_clients(self.num_clients)\n\n            results = self.send_model_and_wait(targets=clients, data=model)\n\n            aggregate_results = self.aggregate(results)\n\n            model = self.update_model(model, aggregate_results)\n\n            self.save_model(model)\n\n        self.info(\"Finished FedAvg.\")\n","highlighted_lines":"7,17,23"},{"id":"job-lightning","type":"job","framework":"lightning","title":"Job Code (fedavg_cifar10_lightning_job.py)","description":"Lastly we construct the job with our 'cifar10_lightning_fl.py' client script and 'FedAvg' server controller. The BaseFedJob automatically configures components for model persistence, model selection, and TensorBoard streaming. We then run the job with the FL simulator.","code":"\nfrom cifar10_lightning_fl import LitNet\n\nfrom nvflare.app_common.workflows.fedavg import FedAvg\nfrom nvflare.app_opt.pt.job_config.base_fed_job import BaseFedJob\nfrom nvflare.job_config.script_runner import ScriptRunner\n\nif __name__ == \"__main__\":\n    n_clients = 2\n    num_rounds = 2\n    train_script = \"cifar10_lightning_fl.py\"\n\n    # Create BaseFedJob with initial model\n    job = BaseFedJob(\n      name=\"cifar10_lightning_fedavg\",\n      initial_model=LitNet(),\n    )\n\n    # Define the controller and send to server\n    controller = FedAvg(\n        num_clients=n_clients,\n        num_rounds=num_rounds,\n    )\n    job.to_server(controller)\n\n    # Add clients\n    for i in range(n_clients):\n        runner = ScriptRunner(script=train_script)\n        job.to(runner, f\"site-{i}\")\n\n    # job.export_job(\"/tmp/nvflare/jobs/job_config\")\n    job.simulator_run(\"/tmp/nvflare/jobs/workdir\", gpu=\"0\")\n"},{"id":"run-lightning","type":"run","framework":"lightning","title":"Run the Job","description":"To run the job with the simulator, copy the code and execute the job script, or run in Google Colab. Alternatively, export the job to a configuration and run the job in other modes.","code":"\npython3 fedavg_cifar10_lightning_job.py\n"}]},{"id":"tensorflow","colab_link":"https://colab.research.google.com/github/NVIDIA/NVFlare/blob/2.5/examples/getting_started/tf/nvflare_tf_getting_started.ipynb","github_link":"https://github.com/NVIDIA/NVFlare/blob/2.5/examples/getting_started/tf/nvflare_tf_getting_started.ipynb","sections":[{"id":"install-tensorflow","type":"install","framework":"tensorflow","title":"Installation","description":"Install the required TensorFlow dependencies for this example.","code":"\npip install nvflare tensorflow\n"},{"id":"client-tensorflow","type":"client","framework":"tensorflow","title":"Client Code (cifar10_tf_fl.py)","description":"We use the Client API to convert the centralized training TensorFlow code into federated learning code with only a few lines of changes highlighted below. Essentially the client will receive a model from NVIDIA FLARE, perform local training and validation, and then send the model back.","code":"\nfrom tensorflow.keras import datasets\nfrom tensorflow.keras import Model, layers, losses\n\n\nclass TFNet(Model):\n    def __init__(self, input_shape):\n        super().__init__()\n        self._input_shape = input_shape  # Required to get constructor arguments in job config\n        self.conv1 = layers.Conv2D(6, 5, activation=\"relu\")\n        self.pool = layers.MaxPooling2D((2, 2), 2)\n        self.conv2 = layers.Conv2D(16, 5, activation=\"relu\")\n        self.flatten = layers.Flatten()\n        self.fc1 = layers.Dense(120, activation=\"relu\")\n        self.fc2 = layers.Dense(84, activation=\"relu\")\n        self.fc3 = layers.Dense(10)\n        loss_fn = losses.SparseCategoricalCrossentropy(from_logits=True)\n        self.compile(optimizer=\"sgd\", loss=loss_fn, metrics=[\"accuracy\"])\n        self.build(input_shape)\n\n    def call(self, x):\n        x = self.pool(self.conv1(x))\n        x = self.pool(self.conv2(x))\n        x = self.flatten(x)\n        x = self.fc1(x)\n        x = self.fc2(x)\n        x = self.fc3(x)\n        return x\n\n\n# (1) import nvflare client API\nimport nvflare.client as flare\n\nPATH = \"./tf_model.ckpt\"\n\n\ndef main():\n    (train_images, train_labels), (test_images, test_labels) = datasets.cifar10.load_data()\n\n    # Normalize pixel values to be between 0 and 1\n    train_images, test_images = train_images / 255.0, test_images / 255.0\n\n    model = TFNet(input_shape=(None, 32, 32, 3))\n    model.summary()\n\n    # (2) initializes NVFlare client API\n    flare.init()\n\n    # (3) gets FLModel from NVFlare\n    while flare.is_running():\n        input_model = flare.receive()\n        print(f\"current_round={input_model.current_round}\")\n\n        # (optional) print system info\n        system_info = flare.system_info()\n        print(f\"NVFlare system info: {system_info}\")\n\n        # (4) loads model from NVFlare\n        for k, v in input_model.params.items():\n            model.get_layer(k).set_weights(v)\n\n        # (5) evaluate aggregated/received model\n        _, test_global_acc = model.evaluate(test_images, test_labels, verbose=2)\n        print(\n            f\"Accuracy of the received model on round {input_model.current_round} on the 10000 test images: \n            {test_global_acc * 100} %\"\n        )\n\n        model.fit(train_images, train_labels, epochs=1, validation_data=(test_images, test_labels))\n\n        print(\"Finished Training\")\n\n        model.save_weights(PATH)\n\n        _, test_acc = model.evaluate(test_images, test_labels, verbose=2)\n        print(f\"Accuracy of the model on the 10000 test images: {test_acc * 100} %\")\n\n        # (6) construct trained FL model (A dict of {layer name: layer weights} from the keras model)\n        output_model = flare.FLModel(\n            params={layer.name: layer.get_weights() for layer in model.layers},\n            metrics={\"accuracy\": test_global_acc}\n        )\n        # (7) send model back to NVFlare\n        flare.send(output_model)\n\n\nif __name__ == \"__main__\":\n    main()\n","highlighted_lines":"31,46,49,50,78-81,83"},{"id":"server-tensorflow","type":"server","framework":"tensorflow","title":"Server Code (fedavg.py)","description":"The ModelController API is used to write a federated routine with mechanisms for distributing and receiving models from clients. Here we implement the basic FedAvg algorithm using some helper functions from BaseFedAvg..","code":"\nfrom nvflare.app_common.workflows.base_fedavg import BaseFedAvg\n\nclass FedAvg(BaseFedAvg):\n    def run(self) -> None:\n        self.info(\"Start FedAvg.\")\n\n        model = self.load_model()\n        model.start_round = self.start_round\n        model.total_rounds = self.num_rounds\n\n        for self.current_round in range(self.start_round, self.start_round + self.num_rounds):\n            self.info(f\"Round {self.current_round} started.\")\n            model.current_round = self.current_round\n\n            clients = self.sample_clients(self.num_clients)\n\n            results = self.send_model_and_wait(targets=clients, data=model)\n\n            aggregate_results = self.aggregate(results)\n\n            model = self.update_model(model, aggregate_results)\n\n            self.save_model(model)\n\n        self.info(\"Finished FedAvg.\")\n","highlighted_lines":"7,17,23"},{"id":"job-tensorflow","type":"job","framework":"tensorflow","title":"Job Code (fedavg_cifar10_tf_job.py)","description":"Lastly we construct the job with our 'cifar10_tf_fl.py' client script and 'FedAvg' server controller. The BaseFedJob automatically configures components for model persistence, model selection, and TensorBoard streaming. We then run the job with the FL simulator.","code":"\nfrom cifar10_tf_fl import TFNet\n\nfrom nvflare.app_common.workflows.fedavg import FedAvg\nfrom nvflare.app_opt.tf.job_config.base_fed_job import BaseFedJob\nfrom nvflare.job_config.script_runner import FrameworkType, ScriptRunner\n\nif __name__ == \"__main__\":\n    n_clients = 2\n    num_rounds = 2\n    train_script = \"cifar10_tf_fl.py\"\n\n    # Create BaseFedJob with initial model\n    job = BaseFedJob(\n      name=\"cifar10_tf_fedavg\",\n      initial_model=TFNet(input_shape=(None, 32, 32, 3)),\n    )\n\n    # Define the controller and send to server\n    controller = FedAvg(\n        num_clients=n_clients,\n        num_rounds=num_rounds,\n    )\n    job.to_server(controller)\n\n    # Add clients\n    for i in range(n_clients):\n        runner = ScriptRunner(\n          script=train_script,\n          framework=FrameworkType.TENSORFLOW,\n        )\n        job.to(runner, f\"site-{i}\")\n\n    # job.export_job(\"/tmp/nvflare/jobs/job_config\")\n    job.simulator_run(\"/tmp/nvflare/jobs/workdir\", gpu=\"0\")\n"},{"id":"run-tensorflow","type":"run","framework":"tensorflow","title":"Run the Job","description":"To run the job with the simulator, copy the code and execute the job script, or run in Google Colab. Alternatively, export the job to a configuration and run the job in other modes.","code":"\npython3 fedavg_cifar10_tf_job.py\n"}]}];
+</span> </button> </li> </ul> <!-- Client wrapper --> <div id="client-wrapper" class="py-4 text-left overflow-x-auto"></div> <!-- Server wrapper --> <div id="server-wrapper" class="py-4 text-left overflow-x-auto hidden"></div> <!-- Job wrapper --> <div id="job-wrapper" class="py-4 text-left overflow-x-auto hidden"></div> </div> <!-- Run wrapper --> <div id="run-wrapper" class="mx-auto max-w-5xl py-4 text-left"></div> </div> </div>     <script type="module">const frameworks = [{"id":"pytorch","colab_link":"https://colab.research.google.com/github/NVIDIA/NVFlare/blob/2.5/examples/getting_started/pt/nvflare_pt_getting_started.ipynb","github_link":"https://github.com/NVIDIA/NVFlare/blob/2.5/examples/getting_started/pt/nvflare_pt_getting_started.ipynb","sections":[{"id":"install-pytorch","type":"install","framework":"pytorch","title":"Installation","description":"Install the required PyTorch dependencies for this example.","code":"\npip install nvflare torch torchvision\n"},{"id":"client-pytorch","type":"client","framework":"pytorch","title":"Client Code (cifar10_pt_fl.py)","description":"We use the Client API to convert the centralized training PyTorch code into federated learning code with only a few lines of changes highlighted below. Essentially the client will receive a model from NVIDIA FLARE, perform local training and validation, and then send the model back.","code":"\nimport torch\nimport torch.nn as nn\nimport torch.optim as optim\nimport torchvision\nimport torchvision.transforms as transforms\n\n\nclass Net(nn.Module):\n    def __init__(self):\n        super().__init__()\n        self.conv1 = nn.Conv2d(3, 6, 5)\n        self.pool = nn.MaxPool2d(2, 2)\n        self.conv2 = nn.Conv2d(6, 16, 5)\n        self.fc1 = nn.Linear(16 * 5 * 5, 120)\n        self.fc2 = nn.Linear(120, 84)\n        self.fc3 = nn.Linear(84, 10)\n\n    def forward(self, x):\n        x = self.pool(F.relu(self.conv1(x)))\n        x = self.pool(F.relu(self.conv2(x)))\n        x = torch.flatten(x, 1) # flatten all dimensions except batch\n        x = F.relu(self.fc1(x))\n        x = F.relu(self.fc2(x))\n        x = self.fc3(x)\n        return x\n\n\n# (1) import nvflare client API\nimport nvflare.client as flare\n\n# (optional) metrics\nfrom nvflare.client.tracking import SummaryWriter\n\n# (optional) set a fix place so we don't need to download everytime\nDATASET_PATH = \"/tmp/nvflare/data\"\n# If available, we use GPU to speed things up.\nDEVICE = \"cuda\" if torch.cuda.is_available() else \"cpu\"\n\n\ndef main():\n    transform = transforms.Compose([\n      transforms.ToTensor(),\n      transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))\n    ])\n\n    batch_size = 4\n    epochs = 2\n\n    trainset = torchvision.datasets.CIFAR10(root=DATASET_PATH, train=True, download=True, transform=transform)\n    trainloader = torch.utils.data.DataLoader(trainset, batch_size=batch_size, shuffle=True, num_workers=2)\n\n    testset = torchvision.datasets.CIFAR10(root=DATASET_PATH, train=False, download=True, transform=transform)\n    testloader = torch.utils.data.DataLoader(testset, batch_size=batch_size, shuffle=False, num_workers=2)\n\n    net = Net()\n\n    # (2) initializes NVFlare client API\n    flare.init()\n\n    summary_writer = SummaryWriter()\n    while flare.is_running():\n        # (3) receives FLModel from NVFlare\n        input_model = flare.receive()\n        print(f\"current_round={input_model.current_round}\")\n\n        # (4) loads model from NVFlare\n        net.load_state_dict(input_model.params)\n\n        criterion = nn.CrossEntropyLoss()\n        optimizer = optim.SGD(net.parameters(), lr=0.001, momentum=0.9)\n\n        # (optional) use GPU to speed things up\n        net.to(DEVICE)\n        # (optional) calculate total steps\n        steps = epochs * len(trainloader)\n        for epoch in range(epochs):  # loop over the dataset multiple times\n\n            running_loss = 0.0\n            for i, data in enumerate(trainloader, 0):\n                # get the inputs; data is a list of [inputs, labels]\n                # (optional) use GPU to speed things up\n                inputs, labels = data[0].to(DEVICE), data[1].to(DEVICE)\n\n                # zero the parameter gradients\n                optimizer.zero_grad()\n\n                # forward + backward + optimize\n                outputs = net(inputs)\n                loss = criterion(outputs, labels)\n                loss.backward()\n                optimizer.step()\n\n                # print statistics\n                running_loss += loss.item()\n                if i % 2000 == 1999:  # print every 2000 mini-batches\n                    print(f\"[{epoch + 1}, {i + 1:5d}] loss: {running_loss / 2000:.3f}\")\n                    global_step = input_model.current_round * steps + epoch * len(trainloader) + i\n\n                    summary_writer.add_scalar(\n                        tag=\"loss_for_each_batch\",\n                        scalar=running_loss, \n                        global_step=global_step\n                    )\n                    running_loss = 0.0\n\n        print(\"Finished Training\")\n\n        PATH = \"./cifar_net.pth\"\n        torch.save(net.state_dict(), PATH)\n\n        # (5) wraps evaluation logic into a method to re-use for\n        #       evaluation on both trained and received model\n        def evaluate(input_weights):\n            net = Net()\n            net.load_state_dict(input_weights)\n            # (optional) use GPU to speed things up\n            net.to(DEVICE)\n\n            correct = 0\n            total = 0\n            # since we're not training, we don't need to calculate the gradients for our outputs\n            with torch.no_grad():\n                for data in testloader:\n                    # (optional) use GPU to speed things up\n                    images, labels = data[0].to(DEVICE), data[1].to(DEVICE)\n                    # calculate outputs by running images through the network\n                    outputs = net(images)\n                    # the class with the highest energy is what we choose as prediction\n                    _, predicted = torch.max(outputs.data, 1)\n                    total += labels.size(0)\n                    correct += (predicted == labels).sum().item()\n\n            print(f\"Accuracy of the network on the 10000 test images: {100 * correct // total} %\")\n            return 100 * correct // total\n\n        # (6) evaluate on received model for model selection\n        accuracy = evaluate(input_model.params)\n        # (7) construct trained FL model\n        output_model = flare.FLModel(\n            params=net.cpu().state_dict(),\n            metrics={\"accuracy\": accuracy},\n            meta={\"NUM_STEPS_CURRENT_ROUND\": steps},\n        )\n        # (8) send model back to NVFlare\n        flare.send(output_model)\n\n\nif __name__ == \"__main__\":\n    main()","highlighted_lines":"29,58,61,63,139-143,145"},{"id":"server-pytorch","type":"server","framework":"pytorch","title":"Server Code (fedavg.py)","description":"The ModelController API is used to write a federated routine with mechanisms for distributing and receiving models from clients. Here we implement the basic FedAvg algorithm using some helper functions from BaseFedAvg.","code":"\nfrom nvflare.app_common.workflows.base_fedavg import BaseFedAvg\n\nclass FedAvg(BaseFedAvg):\n    def run(self) -> None:\n        self.info(\"Start FedAvg.\")\n\n        model = self.load_model()\n        model.start_round = self.start_round\n        model.total_rounds = self.num_rounds\n\n        for self.current_round in range(self.start_round, self.start_round + self.num_rounds):\n            self.info(f\"Round {self.current_round} started.\")\n            model.current_round = self.current_round\n\n            clients = self.sample_clients(self.num_clients)\n\n            results = self.send_model_and_wait(targets=clients, data=model)\n\n            aggregate_results = self.aggregate(results)\n\n            model = self.update_model(model, aggregate_results)\n\n            self.save_model(model)\n\n        self.info(\"Finished FedAvg.\")\n","highlighted_lines":"7,17,23"},{"id":"job-pytorch","type":"job","framework":"pytorch","title":"Job Code (fedavg_cifar10_pt_job.py)","description":"Lastly we construct the job with our 'cifar10_pt_fl.py' client script and 'FedAvg' server controller. The BaseFedJob automatically configures components for model persistence, model selection, and TensorBoard streaming. We then run the job with the FL simulator.","code":"\nfrom cifar10_pt_fl import Net\n\nfrom nvflare.app_common.workflows.fedavg import FedAvg\nfrom nvflare.app_opt.pt.job_config.base_fed_job import BaseFedJob\nfrom nvflare.job_config.script_runner import ScriptRunner\n\nif __name__ == \"__main__\":\n    n_clients = 2\n    num_rounds = 2\n    train_script = \"cifar10_pt_fl.py\"\n\n    # Create BaseFedJob with initial model\n    job = BaseFedJob(\n      name=\"cifar10_pt_fedavg\",\n      initial_model=Net(),\n    )\n\n    # Define the controller and send to server\n    controller = FedAvg(\n        num_clients=n_clients,\n        num_rounds=num_rounds,\n    )\n    job.to_server(controller)\n\n    # Add clients\n    for i in range(n_clients):\n        runner = ScriptRunner(script=train_script)\n        job.to(runner, f\"site-{i}\")\n\n    # job.export_job(\"/tmp/nvflare/jobs/job_config\")\n    job.simulator_run(\"/tmp/nvflare/jobs/workdir\", gpu=\"0\")\n"},{"id":"run-pytorch","type":"run","framework":"pytorch","title":"Run the Job","description":"To run the job with the simulator, copy the code and execute the job script, or run in Google Colab. Alternatively, export the job to a configuration and run the job in other modes.","code":"\npython3 fedavg_cifar10_pt_job.py\n"}]},{"id":"lightning","colab_link":"https://colab.research.google.com/github/NVIDIA/NVFlare/blob/2.5/examples/getting_started/pt/nvflare_lightning_getting_started.ipynb","github_link":"https://github.com/NVIDIA/NVFlare/blob/2.5/examples/getting_started/pt/nvflare_lightning_getting_started.ipynb","sections":[{"id":"install-lightning","type":"install","framework":"lightning","title":"Installation","description":"Install the required PyTorch Lightning dependencies for this example.","code":"\npip install nvflare pytorch_lightning\n"},{"id":"client-lightning","type":"client","framework":"lightning","title":"Client Code (cifar10_lightning_fl.py)","description":"We use the Client API to convert the centralized training PyTorch Lightning code into federated learning code with only a few lines of changes highlighted below. Essentially the client will receive a model from NVIDIA FLARE, perform local training and validation, and then send the model back.","code":"\nimport torch\nimport torch.nn.functional as F\nimport torch.optim as optim\nimport torchvision\nimport torchvision.transforms as transforms\nfrom pytorch_lightning import LightningDataModule, LightningModule Trainer, seed_everything\nfrom torch.utils.data import DataLoader, random_split\nfrom torchmetrics import Accuracy\n\n\nclass LitNet(LightningModule):\n    def __init__(self):\n        super().__init__()\n        self.save_hyperparameters()\n        self.model = Net()\n        self.train_acc = Accuracy(task=\"multiclass\", num_classes=NUM_CLASSES)\n        self.valid_acc = Accuracy(task=\"multiclass\", num_classes=NUM_CLASSES)\n        # (optional) pass additional information via self.__fl_meta__\n        self.__fl_meta__ = {}\n\n    def forward(self, x):\n        out = self.model(x)\n        return out\n\n    def training_step(self, batch, batch_idx):\n        x, labels = batch\n        outputs = self(x)\n        loss = criterion(outputs, labels)\n        self.train_acc(outputs, labels)\n        self.log(\"train_loss\", loss)\n        self.log(\"train_acc\", self.train_acc, on_step=True, on_epoch=False)\n        return loss\n\n    def evaluate(self, batch, stage=None):\n        x, labels = batch\n        outputs = self(x)\n        loss = criterion(outputs, labels)\n        self.valid_acc(outputs, labels)\n\n        if stage:\n            self.log(f\"{stage}_loss\", loss)\n            self.log(f\"{stage}_acc\", self.valid_acc, on_step=True, on_epoch=True)\n        return outputs\n\n    def validation_step(self, batch, batch_idx):\n        self.evaluate(batch, \"val\")\n\n    def test_step(self, batch, batch_idx):\n        self.evaluate(batch, \"test\")\n\n    def predict_step(self, batch: Any, batch_idx: int, dataloader_idx: int = 0) -> Any:\n        return self.evaluate(batch)\n\n    def configure_optimizers(self):\n        optimizer = optim.SGD(self.parameters(), lr=0.001, momentum=0.9)\n        return {\"optimizer\": optimizer}\n\n\nclass CIFAR10DataModule(LightningDataModule):\n    def __init__(self, data_dir: str = DATASET_PATH, batch_size: int = BATCH_SIZE):\n        super().__init__()\n        self.data_dir = data_dir\n        self.batch_size = batch_size\n\n    def prepare_data(self):\n        torchvision.datasets.CIFAR10(root=self.data_dir, train=True, download=True, transform=transform)\n        torchvision.datasets.CIFAR10(root=self.data_dir, train=False, download=True, transform=transform)\n\n    def setup(self, stage: str):\n        # Assign train/val datasets for use in dataloaders\n        if stage == \"fit\" or stage == \"validate\":\n            cifar_full = torchvision.datasets.CIFAR10(\n                root=self.data_dir, train=True, download=False, transform=transform\n            )\n            self.cifar_train, self.cifar_val = random_split(cifar_full, [0.8, 0.2])\n\n        # Assign test dataset for use in dataloader(s)\n        if stage == \"test\" or stage == \"predict\":\n            self.cifar_test = torchvision.datasets.CIFAR10(\n                root=self.data_dir, train=False, download=False, transform=transform\n            )\n\n    def train_dataloader(self):\n        return DataLoader(self.cifar_train, batch_size=self.batch_size)\n\n    def val_dataloader(self):\n        return DataLoader(self.cifar_val, batch_size=self.batch_size)\n\n    def test_dataloader(self):\n        return DataLoader(self.cifar_test, batch_size=self.batch_size)\n\n    def predict_dataloader(self):\n        return DataLoader(self.cifar_test, batch_size=self.batch_size)\n\n\n# (1) import nvflare lightning client API\nimport nvflare.client.lightning as flare\n\nseed_everything(7)\n\n\nDATASET_PATH = \"/tmp/nvflare/data\"\nBATCH_SIZE = 4\n\ntransform = transforms.Compose([transforms.ToTensor(), transforms.Normalize((0.5, 0.5, 0.5), (0.5, 0.5, 0.5))])\n\ndef main():\n    model = LitNet()\n    cifar10_dm = CIFAR10DataModule()\n    if torch.cuda.is_available():\n        trainer = Trainer(max_epochs=1, accelerator=\"gpu\", devices=1 if torch.cuda.is_available() else None)\n    else:\n        trainer = Trainer(max_epochs=1, devices=None)\n\n    # (2) patch the lightning trainer\n    flare.patch(trainer)\n\n    while flare.is_running():\n        # (3) receives FLModel from NVFlare\n        # Note that we don't need to pass this input_model to trainer\n        # because after flare.patch the trainer.fit/validate will get the\n        # global model internally\n        input_model = flare.receive()\n        print(f\"\\n[Current Round={input_model.current_round}, Site = {flare.get_site_name()}]\\n\")\n\n        # (4) evaluate the current global model to allow server-side model selection\n        print(\"--- validate global model ---\")\n        trainer.validate(model, datamodule=cifar10_dm)\n\n        # perform local training starting with the received global model\n        print(\"--- train new model ---\")\n        trainer.fit(model, datamodule=cifar10_dm)\n\n        # test local model\n        print(\"--- test new model ---\")\n        trainer.test(ckpt_path=\"best\", datamodule=cifar10_dm)\n\n        # get predictions\n        print(\"--- prediction with new best model ---\")\n        trainer.predict(ckpt_path=\"best\", datamodule=cifar10_dm)\n\n\nif __name__ == \"__main__\":\n    main()\n","highlighted_lines":"97,116,118,123"},{"id":"server-lightning","type":"server","framework":"lightning","title":"Server Code (fedavg.py)","description":"The ModelController API is used to write a federated routine with mechanisms for distributing and receiving models from clients. Here we implement the basic FedAvg algorithm using some helper functions from BaseFedAvg.","code":"\nfrom nvflare.app_common.workflows.base_fedavg import BaseFedAvg\n\nclass FedAvg(BaseFedAvg):\n    def run(self) -> None:\n        self.info(\"Start FedAvg.\")\n\n        model = self.load_model()\n        model.start_round = self.start_round\n        model.total_rounds = self.num_rounds\n\n        for self.current_round in range(self.start_round, self.start_round + self.num_rounds):\n            self.info(f\"Round {self.current_round} started.\")\n            model.current_round = self.current_round\n\n            clients = self.sample_clients(self.num_clients)\n\n            results = self.send_model_and_wait(targets=clients, data=model)\n\n            aggregate_results = self.aggregate(results)\n\n            model = self.update_model(model, aggregate_results)\n\n            self.save_model(model)\n\n        self.info(\"Finished FedAvg.\")\n","highlighted_lines":"7,17,23"},{"id":"job-lightning","type":"job","framework":"lightning","title":"Job Code (fedavg_cifar10_lightning_job.py)","description":"Lastly we construct the job with our 'cifar10_lightning_fl.py' client script and 'FedAvg' server controller. The BaseFedJob automatically configures components for model persistence, model selection, and TensorBoard streaming. We then run the job with the FL simulator.","code":"\nfrom cifar10_lightning_fl import LitNet\n\nfrom nvflare.app_common.workflows.fedavg import FedAvg\nfrom nvflare.app_opt.pt.job_config.base_fed_job import BaseFedJob\nfrom nvflare.job_config.script_runner import ScriptRunner\n\nif __name__ == \"__main__\":\n    n_clients = 2\n    num_rounds = 2\n    train_script = \"cifar10_lightning_fl.py\"\n\n    # Create BaseFedJob with initial model\n    job = BaseFedJob(\n      name=\"cifar10_lightning_fedavg\",\n      initial_model=LitNet(),\n    )\n\n    # Define the controller and send to server\n    controller = FedAvg(\n        num_clients=n_clients,\n        num_rounds=num_rounds,\n    )\n    job.to_server(controller)\n\n    # Add clients\n    for i in range(n_clients):\n        runner = ScriptRunner(script=train_script)\n        job.to(runner, f\"site-{i}\")\n\n    # job.export_job(\"/tmp/nvflare/jobs/job_config\")\n    job.simulator_run(\"/tmp/nvflare/jobs/workdir\", gpu=\"0\")\n"},{"id":"run-lightning","type":"run","framework":"lightning","title":"Run the Job","description":"To run the job with the simulator, copy the code and execute the job script, or run in Google Colab. Alternatively, export the job to a configuration and run the job in other modes.","code":"\npython3 fedavg_cifar10_lightning_job.py\n"}]},{"id":"tensorflow","colab_link":"https://colab.research.google.com/github/NVIDIA/NVFlare/blob/2.5/examples/getting_started/tf/nvflare_tf_getting_started.ipynb","github_link":"https://github.com/NVIDIA/NVFlare/blob/2.5/examples/getting_started/tf/nvflare_tf_getting_started.ipynb","sections":[{"id":"install-tensorflow","type":"install","framework":"tensorflow","title":"Installation","description":"Install the required TensorFlow dependencies for this example.","code":"\npip install nvflare tensorflow\n"},{"id":"client-tensorflow","type":"client","framework":"tensorflow","title":"Client Code (cifar10_tf_fl.py)","description":"We use the Client API to convert the centralized training TensorFlow code into federated learning code with only a few lines of changes highlighted below. Essentially the client will receive a model from NVIDIA FLARE, perform local training and validation, and then send the model back.","code":"\nfrom tensorflow.keras import datasets\nfrom tensorflow.keras import Model, layers, losses\n\n\nclass TFNet(Model):\n    def __init__(self, input_shape):\n        super().__init__()\n        self._input_shape = input_shape  # Required to get constructor arguments in job config\n        self.conv1 = layers.Conv2D(6, 5, activation=\"relu\")\n        self.pool = layers.MaxPooling2D((2, 2), 2)\n        self.conv2 = layers.Conv2D(16, 5, activation=\"relu\")\n        self.flatten = layers.Flatten()\n        self.fc1 = layers.Dense(120, activation=\"relu\")\n        self.fc2 = layers.Dense(84, activation=\"relu\")\n        self.fc3 = layers.Dense(10)\n        loss_fn = losses.SparseCategoricalCrossentropy(from_logits=True)\n        self.compile(optimizer=\"sgd\", loss=loss_fn, metrics=[\"accuracy\"])\n        self.build(input_shape)\n\n    def call(self, x):\n        x = self.pool(self.conv1(x))\n        x = self.pool(self.conv2(x))\n        x = self.flatten(x)\n        x = self.fc1(x)\n        x = self.fc2(x)\n        x = self.fc3(x)\n        return x\n\n\n# (1) import nvflare client API\nimport nvflare.client as flare\n\nPATH = \"./tf_model.ckpt\"\n\n\ndef main():\n    (train_images, train_labels), (test_images, test_labels) = datasets.cifar10.load_data()\n\n    # Normalize pixel values to be between 0 and 1\n    train_images, test_images = train_images / 255.0, test_images / 255.0\n\n    model = TFNet(input_shape=(None, 32, 32, 3))\n    model.summary()\n\n    # (2) initializes NVFlare client API\n    flare.init()\n\n    # (3) gets FLModel from NVFlare\n    while flare.is_running():\n        input_model = flare.receive()\n        print(f\"current_round={input_model.current_round}\")\n\n        # (optional) print system info\n        system_info = flare.system_info()\n        print(f\"NVFlare system info: {system_info}\")\n\n        # (4) loads model from NVFlare\n        for k, v in input_model.params.items():\n            model.get_layer(k).set_weights(v)\n\n        # (5) evaluate aggregated/received model\n        _, test_global_acc = model.evaluate(test_images, test_labels, verbose=2)\n        print(\n            f\"Accuracy of the received model on round {input_model.current_round} on the 10000 test images: \n            {test_global_acc * 100} %\"\n        )\n\n        model.fit(train_images, train_labels, epochs=1, validation_data=(test_images, test_labels))\n\n        print(\"Finished Training\")\n\n        model.save_weights(PATH)\n\n        _, test_acc = model.evaluate(test_images, test_labels, verbose=2)\n        print(f\"Accuracy of the model on the 10000 test images: {test_acc * 100} %\")\n\n        # (6) construct trained FL model (A dict of {layer name: layer weights} from the keras model)\n        output_model = flare.FLModel(\n            params={layer.name: layer.get_weights() for layer in model.layers},\n            metrics={\"accuracy\": test_global_acc}\n        )\n        # (7) send model back to NVFlare\n        flare.send(output_model)\n\n\nif __name__ == \"__main__\":\n    main()\n","highlighted_lines":"31,46,49,50,78-81,83"},{"id":"server-tensorflow","type":"server","framework":"tensorflow","title":"Server Code (fedavg.py)","description":"The ModelController API is used to write a federated routine with mechanisms for distributing and receiving models from clients. Here we implement the basic FedAvg algorithm using some helper functions from BaseFedAvg.","code":"\nfrom nvflare.app_common.workflows.base_fedavg import BaseFedAvg\n\nclass FedAvg(BaseFedAvg):\n    def run(self) -> None:\n        self.info(\"Start FedAvg.\")\n\n        model = self.load_model()\n        model.start_round = self.start_round\n        model.total_rounds = self.num_rounds\n\n        for self.current_round in range(self.start_round, self.start_round + self.num_rounds):\n            self.info(f\"Round {self.current_round} started.\")\n            model.current_round = self.current_round\n\n            clients = self.sample_clients(self.num_clients)\n\n            results = self.send_model_and_wait(targets=clients, data=model)\n\n            aggregate_results = self.aggregate(results)\n\n            model = self.update_model(model, aggregate_results)\n\n            self.save_model(model)\n\n        self.info(\"Finished FedAvg.\")\n","highlighted_lines":"7,17,23"},{"id":"job-tensorflow","type":"job","framework":"tensorflow","title":"Job Code (fedavg_cifar10_tf_job.py)","description":"Lastly we construct the job with our 'cifar10_tf_fl.py' client script and 'FedAvg' server controller. The BaseFedJob automatically configures components for model persistence and model selection. We then run the job with the FL simulator.","code":"\nfrom cifar10_tf_fl import TFNet\n\nfrom nvflare.app_common.workflows.fedavg import FedAvg\nfrom nvflare.app_opt.tf.job_config.base_fed_job import BaseFedJob\nfrom nvflare.job_config.script_runner import FrameworkType, ScriptRunner\n\nif __name__ == \"__main__\":\n    n_clients = 2\n    num_rounds = 2\n    train_script = \"cifar10_tf_fl.py\"\n\n    # Create BaseFedJob with initial model\n    job = BaseFedJob(\n      name=\"cifar10_tf_fedavg\",\n      initial_model=TFNet(input_shape=(None, 32, 32, 3)),\n    )\n\n    # Define the controller and send to server\n    controller = FedAvg(\n        num_clients=n_clients,\n        num_rounds=num_rounds,\n    )\n    job.to_server(controller)\n\n    # Add clients\n    for i in range(n_clients):\n        runner = ScriptRunner(\n          script=train_script,\n          framework=FrameworkType.TENSORFLOW,\n        )\n        job.to(runner, f\"site-{i}\")\n\n    # job.export_job(\"/tmp/nvflare/jobs/job_config\")\n    job.simulator_run(\"/tmp/nvflare/jobs/workdir\", gpu=\"0\")\n"},{"id":"run-tensorflow","type":"run","framework":"tensorflow","title":"Run the Job","description":"To run the job with the simulator, copy the code and execute the job script, or run in Google Colab. Alternatively, export the job to a configuration and run the job in other modes.","code":"\npython3 fedavg_cifar10_tf_job.py\n"}]}];
 
   // @ts-nocheck
   const frameworkLabels = document.querySelectorAll("#framework-labels > span");