Release v0.16.0-rc1

Major changes and bug fixes

• Targeted for the PlaidML backend release
• nGraph version updated to v0.23.0-rc.3
• Fixed broken Mac build
• Added batch size support for C++ example
• Fixed issues in distributed training

Dependency information

TensorFlow v1.14.0
nGraph v0.23.0-rc.3

Release v0.16.0-rc0

Major changes and bug fixes

• Targeted for the Intel GPU backend release
• nGraph version updated to v0.22.0
• TensorFlow version updated to v1.14.0
• Capability to add backend configurations via rewriter config.
• New build option: --use_prebuilt_tensorflow this cuts down the build time sigfinicantly

Dependency information

TensorFlow v1.14.0
nGraph v0.22.0

Release v0.14.2-rc0

Major changes and bug fixes

• nGraph version updated to v0.21.0

Dependency information

TensorFlow v1.13.1
nGraph v0.21.0

Release v0.15.0-rc1

Dependency information

TensorFlow v1.14.0-rc0
nGraph v0.20.0-rc.1

Release v0.15.0-rc0

Dependency information

TensorFlow v1.14.0-rc0
nGraph v0.20.0-rc.1

Release v0.14.1

Major changes and bug fixes

• Added Bazel build script for building CPU backend: bazel build @ngraph//:libcpu_backend.so
• Added option to specify ngraph backend in tf2ngraph: —ngbackend <backend name>
• nGraph version updated to v0.20.0-rc.0

Dependency information

TensorFlow v1.13.1
nGraph v0.20.0-rc.0

Release v0.14.0

Major changes and bug fixes

• Upgraded nGraph to v0.19.1
• Capture Pass Checks: Capture Assign and other modifiers only if they operate on a variable and if the attribute validate_shape is true.
• API to disable certain Tensorflow node types - useful for experiments and tests.
• Added new Ops - RsqrtGrad, FusedMatMul, LogSoftMax
• Added tf2ngraph.py - a script that transforms a TensorFlow graph to nGraph offline using Grappler.

Dependency information

TensorFlow v1.13.1
nGraph v0.19.1

Release v0.14.0-rc0

Major changes and bug fixes

• First release from our new repository location
• Upgraded nGraph to v0.19.0
• More grappler based improvements, new op support and bug fixes

Dependency information

TensorFlow r1.13.1
nGraph v0.19.0

Release v0.13.0-rc1

Major changes and bug fixes

1. Upgraded nGraph to v0.19.0

Dependency information

TensorFlow r1.14 Nightly
nGraph v0.19.0

⚠️ Since TensorFlow 1.14 hasn't been published, currently you need to use either the nightly build (tf-nightly==1.14.1.dev20190515) or build TensorFlow 1.14 from source.

Release v0.13.0-rc0

Major changes and bug fixes

1. Upgraded nGraph to v0.19.0-rc.4.
2. Capture Pass Checks: Capture Assign and other modifiers only if they operate on a variable and if the attribute validate_shape is true.
3. Made Grappler as default optimizer. Use update session config API to update the Session Config before using nGraph.
4. API to disable certain Tensorflow node types - useful for experiments and tests.
5. Added new Ops - RsqrtGrad, FusedMatMul, CombinedNMS

Dependency information

TensorFlow r1.14
nGraph v0.19.0-rc.4

⚠️ Since TensorFlow 1.14.0 hasn't been published, currently you need to use either the nightly build (tf-nightly==1.14.1.dev20190515) or build TensorFlow 1.14 from source.