tobe

Have you changed the error message of this problem? @sailor88128 Now it seems that the python client works but the server has not started. So you get the `StatusCode.UNAVAILABLE` problem....

Now you can install the python sdk with `pip install tensorflow-serving-api`.

It seems that you have saved the checkpoint in other network. This script will try to load the checkpoint from the path "./checkpoint" firstly and it throws errors when you...

Can you provide the error log about this?

I think we have changed `key` to `keys` when exporting the model. You can try modify the code and see if it's solved. By the way, I recommend you to...

Hi @laolu1573 , the classes of `tensorflow.serving.*` are generated by the maven protoc plugin. You can run the `mvn clean package` to generate these dependencies automatically.

Thanks @petro-rudenko . We have ty lz4 codec in this scenario and it works with SparkRDMA.

By the way, we got another error after shuffle read when compression was disable. So it is definitely not the issue of the compression implementation.

@petro-rudenko The error message for disable compression was in the description of this issue. The third code snippet of https://github.com/Mellanox/SparkRDMA/issues/34#issue-465161133 .

We have communicated with the exports of FPGA and they said that the initialization of each FPGA process will request and reset some memory which may affect the data of...