Hi Rakesh,

You can use DirectRunner to run Tensorflow on Beam. Please refer
https://www.tensorflow.org/tfx/transform/get_started to learn more about it.
We are working on making Tensorflow available on Flink Runner. You can
build Beam from Master and run the Tensorflow model on it.
Here are general instructions to run Portable FlinkRunner
https://beam.apache.org/contribute/portability/#status
Please add the following parameters "--experiments= worker_threads=100" and
"--experiments=beam_fnapi" to your pipeline arguments.
For reference, here is a sample of TensorFlow Job on FlinkRunner
https://github.com/angoenka/model-analysis/blob/axelmagn-hacks/examples/chicago_taxi/preprocess_flink.sh
Note, There might be some rough edges as the PortableFlinkRunner is still
in development.

Thanks,
Ankur


On Sat, Aug 25, 2018 at 12:58 PM Rakesh Kumar <[email protected]> wrote:

> Hi all,
>
> I would like to know whether Tensor Flow is supported on Flink Runner &
> DirectRunner? If it is not currently supported then what is the plan to
> provide the support.
>
> Thank you
> Rakesh
> --
> Rakesh Kumar
> Software Engineer
> 510-761-1364 |
>
> <https://www.lyft.com/>
>

Reply via email to