Yan Xu

Make TO RUN component don't bind one SQL engine

2

From the current [TO RUN](https://github.com/sql-machine-learning/sqlflow/blob/develop/doc/design/to_run.md) design doc, the component using environment variable as the context, that caused two problems: 1. The component would bind to one SQL engine 1. Can...

Setup a password when launching playground with single-user mode

3

If we want to launch palyground on Google Cloud, Aliyun, or other public clouds provider. Playground exposes the Jupyter Notebook port on public internal, that we should set a password...

Using `kubectl create -f install-sqlflow-multi-users.yaml` to setup a JupyterHub server, for each login user, JupytertHub would setup a MySQL server and an SQLFlow server with Jupyter Notebook server.

Can we remove the parameter server implementation with Python

1

There is two entry point of the parameter server instance, the one is Python [/elasticdl/python/ps](https://github.com/sql-machine-learning/elasticdl/blob/develop/elasticdl/python/ps/main.py), another one is [/elasticdl/go/cmd/elasticdl_ps](https://github.com/sql-machine-learning/elasticdl/blob/develop/elasticdl/go/cmd/elasticdl_ps/main.go). It seems the Python code is useless, can we remove it...

Add some unit tests to check i32 shape

3

In the current codebase, we can build mhlo with `TORCH_MLIR_ENABLE_MHLO_TRUNC_DIMSIZE_TO_I32` to enable the i32 shape, but there are no unit tests to test the i32 feature, we need a way...

Compatible with previous PyTorch version

5

Hi torch-mlir team, I came from the [BladeDISC](https://github.com/alibaba/BladeDISC) repo. We noticed that torch-mlir always follows the latest PyTorch version, and I want to know does torch-mlir team has any plan...

1. use PFS instead of tunnel host. 1. `paddlectl` command logs need to update. 1. use `PipeReader` to read date from hadoop 1. etc...