An open-source framework for machine learning and other computations on decentralized data.
APACHE-2.0 License
Bot releases are hidden (Show)
Published by amlanchakoty 4 months ago
Published by huili0140 5 months ago
tff.learning.models.functional_model_from_keras
.com_github_grpc_grpc
to version 1.50.0
.Published by nicolemitchell 5 months ago
rules_license
to version 0.0.8
.elias_gamma_encode
module.tensorflow_compression
dependency.Published by huili0140 6 months ago
__eq__()
on building blocks
.-march=haswell
) instead ofPublished by amlanchakoty 6 months ago
Literal
to the TFF language, part 2. This change updates theLiteral
.tff.learning.optimizers.build_adafactor
content
, to the Data
proto.check_foo()
methods on building blocks.tff.data
, this symbol is not used.Lambda
s under sequence_*
intrinsics.Published by nicolemitchell 7 months ago
Literal
to the TFF language, part 1.Literal
to the framework package.tff.learning.algorithms.build_weighted_fed_avg_with_optimizer_schedule
totff.learning.models.FunctionalModel
.tff.learning.framework
namespace⚰️.FilePerUserClientData.create_tf_dataset_for_client
could blow up memoryPublished by amlanchakoty 7 months ago
DataType
protobuf object into its own module. Moving theDataType
object into its own module allows DataType
to be used outsideComputation
more easily and prevents a circular dependency betweenComputation
and Array
which both require a DataType
.build_apply_optimizer_finalizer
to allow custom reject updateModelWeights
to allowtf.keras.Model
instances.tff.learning
APIs.tf.RaggedTensor
andtf.SparseTensor
.Full Changelog: https://github.com/tensorflow/federated/compare/v0.73.0...v0.74.0
Published by ZacharyGarrett 7 months ago
tff.learning.algorithms.build_fed_kmeans
supports floating point weights,tff.aggregators
using differential privacy.tff.learning.metrics.finalize_then_sample
andtff.learning.metrics.FinalizeThenSampleFactory
.SequenceType
from tff.federated_computation
tff.learning
algorithms now correctly do not include metrics for clientsPublished by huili0140 8 months ago
tff.backends.xla
.tensorflow-privacy
version to 0.9.0
.type_signature
parameter from the tff.program.ReleaseManager.release
method.Published by nicolemitchell 8 months ago
Published by zcharles8 9 months ago
tff.program.PrefetchingDataSource
due to flakinessinfer_type
with TensorFlow values.tff.aggregators.federated_(min|max)
symbols, please usetff.federated_(min|max)
instead.tff.TensorType
using a tf.dtypes.DType
.tff.check_return_type
.OwnedValueId::INVALID_ID
as a static constexpr.Published by chloekiddon 9 months ago
local_unfinalized_metrics_type
argument totf.TensorSpec
using to_type
, usetensorflow_to_type
instead.tff.TensorType
using a tf.dtypes.DType
.Published by michaelreneer 9 months ago
tff.types.tensorflow_to_type
function to convert structurestff.Type
.tff.types.infer_unplaced_type
.tff.types.ArrayShape
to be defined as a Sequence
not anIterable
, this is because the len
of an tff.types.ArrayShape
is usedtype_signature
parameter for thetff.program.ReleaseManager.release
method.tff.Value.__add__
.tff.Type.check_*()
functions, use isinstance
tff.types.at_clients
and tff.types.at_server
functions, use thetff.FederatedType
constructor instead.tf.data.DatasetSpec
, tf.RaggedTensorSpec
,tf.SparseTensorSpec
using tff.to_type
, usetff.types.tensorflow_to_type
instead.tf.RaggedTensor
and tf.SparseTensor
usinginfer_type
.Published by huili0140 10 months ago
np.dtype
tf.dtypes.Dtype
.tff.program.DelayedReleaseManager
.check_allowed_ops
from the framework
package.check_disallowed_ops
from the framework
package.replace_intrinsics_with_bodies
from the framework
package.get_session_token
from the framework
package.pybind11_bazel
.type_from_tensors
from the framework
package.rules_python
to 0.23.0
.googleapis-common-protos
to version 1.61.0
to work aroundPublished by nicolemitchell 11 months ago
tf.TensorShape
as the shape of atff.TensorType
.SERVER
placed values out of the C++ execution stackPublished by michaelreneer 12 months ago
Update the representation of a tff.TensorType.shape
to be a
tff.types.ArrayShape
instead of tf.TensorShape
.
Updated type_to_py_container``to be able to handle
tff.SequenceTypes` with
an unknown Python type.
tff.structure_from_tensor_type_tree
totff.types.structure_from_tensor_type_tree
.int
as the shape of a tff.TensorType
.Published by huili0140 12 months ago
train_process
to train_process_factory
in vizier program logicgrpc
to v1.59.1
.bazel
to 6.1.0
.tensorflow
to 2.14.0
.numpy
to ~=1.25
.com_google_googletest
to 1.12.1
.tff.program.FileProgramStateManager
removing old programPublished by zcharles8 about 1 year ago
tff.federated_min
and tff.federated_max
intrinsics.get_value()
method to tff.program.SavedModelFileReleaseManager,
for retrieving values that were previously released.tff.program.PeriodicReleaseManager
to release values at regular intervals.DistributeAggregateForm
now skips normalizing the all_equal bit.attrs
to 23.1
.cachetools
to ~=5.3
.dp-accounting
to 0.4.3
.google-vizier
to 0.1.11
.jax
to 0.4.14
.portpicker
to ~=1.6
.tensorflow
to 2.13.0
.tensorflow-model-optimization
to 0.7.5
.tensorflow-privacy
to 0.8.11
.typing-extensions
to ~=4.5.0
.TF_CUDA_VERSION
to 12
.tff.program.Capabilities
enum from the iterator.is_{foo}
functions from tff.Type
s. Users should use isinstance
checks instead.tff.program
package for consistency and clarity.undefined-variable
lint errors.UnicodeDecodeError
in the FedRecon tutorial.Published by ZacharyGarrett about 1 year ago
context
argument fromtff.learning.algorithms.build_personalization_eval_computation
. Now atff.learning.ddp_secure_aggregator
.Published by xiaoyux11 over 1 year ago
dtype
parameter to tff.TensorType
.ThresholdSampling
class.tff.learning.models.ReconstructionModel.from_keras_model_and_variables
,ReconstructionModel
from a Keras model andtff.learning.algorithms.build_fed_recon_eval
to use a statefultff.learning.models.ReconstructionModel.from_keras_model
, whichtff.learning.models.ReconstructionModel.from_keras_model_and_layers
.check_len
,check_callable
, is_dataclass
,is_attrs
, check_subclass
andcheck_not_none
. They are unused or can be replaced by Python type