Update the version to 0.19.0dev

Update the version to 0.18.0

Signed-off-by: Krzysztof Lecki <[email protected]>

Fix problem with TensorFlow and cupy tests (NVIDIA#1568)

- when tensorflow-gpu v1.15.0 is installed in the conda environment
  some strange dependency leads to `ModuleNotFoundError: No module
  named 'prompt_toolkit.formatted_text'` error
- according to jupyter/notebook#4050 installing
  `ipykernel<5.0.0 ipython<7.0.0` helps
- fixes cupy version to 6.6.0 as this the last one supporting python 2.7

Signed-off-by: Janusz Lisiecki <[email protected]>

Update the version to 0.18.0-dev

Signed-off-by: Krzysztof Lecki <[email protected]>

Fix scalar batch handling in arithmetic ops (NVIDIA#1449)

Adjust test for checking if the Tensor should
be considered a scalar - uniform shape
of 1-dim scalar elements.
Allow Scalar batch to be broadcasted in the operation
properly - no offset in tiles for scalar-like data,
same code-path as constants.

Return shape is always a batch - fixed for Constant/Scalar
inputs.

Add C++ unit tests for Scalar Batch.
Test if Pipeline behaves properly when switching between
scalar and non-scalar inputs.

Python test was extended to cover Scalar inputs,
size of some of the tests was reduced.

Signed-off-by: Krzysztof Lecki <[email protected]>

Update the Version to 0.17.0dev

Signed-off-by: Krzysztof Lecki <[email protected]>

Relax check for libnvidia-opticalflow is test script. (NVIDIA#1381)

Relax check for libnvidia-opticalflow is test script. (NVIDIA#1381)

* Check if neither libnvidia-opticalflow.so nor libnvidia-opticalflow.so.1 exists.

Signed-off-by: Michal Zientkiewicz <[email protected]>

Provide prebuilt plugins for manylinux2010 based pip packages (NVIDIA…

…#1346)

Signed-off-by: Joaquin Anton <[email protected]>

Change version to 0.16.0-dev

Signed-off-by: Krzysztof Lecki <[email protected]>