A serverless federated learning library based on flwr
Project description
A Flower (flwr) extension for serverless federated learning.
Technical report (arXiv): Serverless Federated Learning with flwr-serverless
.
Install
pip install flwr-serverless
or
pip install git+https://github.com/kungfuai/flwr_serverless.git
Usage for tensorflow
- Step 1: Create federated
Node
s that use a shared folder to exchange model weights and use a federated strategy (flwr.server.strategy.Strategy
) to control how the weights are aggregated. - Step 2: Create and configure a callback
FlwrFederatedCallback
and use it in thekeras.Model.fit()
.
# Create a FL Node that has a strategy and a shared folder.
from flwr.server.strategy import FedAvg # This is a flwr federated strategy.
from flwr_serverless import AsyncFederatedNode, S3Folder
from flwr_serverless.keras import FlwrFederatedCallback
strategy = FedAvg()
shared_folder = S3Folder(directory="mybucket/experiment1")
node = AsyncFederatedNode(strategy=strategy, shared_folder=shared_folder)
# Create a keras Callback with the FL node.
num_examples_per_epoch = steps_per_epoch * batch_size # number of examples used in each epoch
callback = FlwrFederatedCallback(
node,
num_examples_per_epoch=num_examples_per_epoch,
save_model_before_aggregation=False,
save_model_after_aggregation=False,
)
# Join the federated learning, by fitting the model with the federated callback.
model = keras.Model(...)
model.compile(...)
model.fit(dataset, callbacks=[callback])
flwr_serverless
uses flwr_serverless.SharedFolder
to save model weights and metrics. The logic folder can be backed by a storage backend like S3.
The asynchronous FL node does not wait to sync with other nodes. It takes the latest model weights from other nodes and performs the aggregation according to the specified strategy.
Running experiments
To make it easier to experimemt with different strategies, we provide utility classes like flwr.keras.example.FederatedLearningTestRun
. This allows you to configure the dataset partition, strategy and concurrency. Please use this as an example to develop your own experiments.
To reproduce some experiments reported in the paper, run
python -m experiments.experiment_scripts.exp1_mnist
python -m experiments.experiment_scripts.exp2_cifar10
python -m experiments.experiment_scripts.exp3_wikitext
Each of the above experiments run through a grid search over a large hyperparameter space, with repeated trials using different random seeds. Please edit the script to adjust the number of trials and the hyperparameter search space.
Project details
Release history Release notifications | RSS feed
Download files
Download the file for your platform. If you're not sure which to choose, learn more about installing packages.
Source Distribution
Built Distribution
File details
Details for the file flwr_serverless-0.2.10.tar.gz
.
File metadata
- Download URL: flwr_serverless-0.2.10.tar.gz
- Upload date:
- Size: 18.3 kB
- Tags: Source
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/5.0.0 CPython/3.12.2
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | 93abe0040be8724197d05b1d43608167bc6443bb4da3152e7e2d9a3a192a9460 |
|
MD5 | fcb7d078174379503983c3eab90895a1 |
|
BLAKE2b-256 | a6f3619fb1d7acfa29bb50fd5d259d3ec32c19e2232a0a27464814247923c842 |
File details
Details for the file flwr_serverless-0.2.10-py3-none-any.whl
.
File metadata
- Download URL: flwr_serverless-0.2.10-py3-none-any.whl
- Upload date:
- Size: 24.1 kB
- Tags: Python 3
- Uploaded using Trusted Publishing? No
- Uploaded via: twine/5.0.0 CPython/3.12.2
File hashes
Algorithm | Hash digest | |
---|---|---|
SHA256 | b5eb8b5aa51988f2190cbcc642e24d22783cb2138ff0c7ff57575b5d3cf9e268 |
|
MD5 | 9ae0b21f4657d9d52da967411e221067 |
|
BLAKE2b-256 | 58d9de271b8b8d7ff5a73505290cff48f231022427c5370fa0b87e48d8c84cde |