DirectML PluggableDevice plugin for TensorFlow 2
APACHE-2.0 License
Bot releases are hidden (Show)
Published by maggie1059 over 1 year ago
The Python packages are available as a PyPI release. To download the latest python package automatically, simply pip install tensorflow-directml-plugin
.
CudnnRNNCanonicalToParams
and CudnnRNNParamsToCanonical
Conv2DBackpropFilter
and Conv3DBackpropFilter
float16
support for _FusedConv2D
Published by maggie1059 almost 2 years ago
The Python packages are available as a PyPI release. To download the latest python package automatically, simply pip install tensorflow-directml-plugin
.
tensorflow-cpu==2.10.0
as a hard dependency due to incompatibility with Keras 2.11's default optimizers.BatchNorm
ops when float16 or mixed precision is used.Cast
operation in ReduceMin
and ReduceMax
ops.Published by PatriceVignola almost 2 years ago
The Python packages are available as a PyPI release. To download the latest python package automatically, simply pip install tensorflow-directml-plugin
.
exponential_avg_factor != 1.0
in FusedBatchNorm
int32
kernel registration for Fill
Published by PatriceVignola about 2 years ago
The Python packages are available as a PyPI release. To download the latest python package automatically, simply pip install tensorflow-directml-plugin
.
InTopKV2
when k
is bigger than the size of the axis dimension.Published by PatriceVignola about 2 years ago
The Python packages are available as a PyPI release. To download the latest python package automatically, simply pip install tensorflow-directml-plugin
.
RngSkip
and RngReadAndSkip
operators.StatelessRandomGetKeyCounterAlg
, StatelessRandomGetKeyCounter
and StatelessRandomGetAlg
operators.SparseApplyAdagrad
.StatelessRandomUniformV2
.InTopKV2
.MatrixDiagV3
and MatrixDiagPartV3
.int64
.tensorflow-cpu>=2.10.0
. Users should install the tensorflow-cpu
package instead of tensorflow
or tensorflow-gpu
when using tensorflow-directml-plugin
.int32
support for StridedSlice
.UnsortedSegmentSum
, UnsortedSegmentMax
, UnsortedSegmentMin
and UnsortedSegmentProd
to get rid of device placement errors in transformer models.tensorflow-directml-plugin
GitHub repository.Relu
.int32
support for Pack
.ArgMin
and ArgMax
when the output type was int16
or uint16
.SegmentReduction
.StridedSlice
.Range
to allow values near INT_MAX
.DataFormatDimMap
and DataFormatVecPermute
operators.