Is using deepspeed.launcher.launch a good practice?

[matbun]

I found some SLURM jobscripts written by some colleagues of mine setting up a DeepSpeed cluster by running a command like:

srun python -m deepspeed.launcher.launch \
     --node_rank $SLURM_PROCID \
     --master_addr ${SLURMD_NODENAME}i \
     --master_port 29500 \
     --world_info $WID \
     my_python_script.py $MY_PYTHON_ARGS\
    --deepspeed_mpi --deepspeed_config DS_config.json

However, on the docs page I could not find this launch method. Rather, it is suggested to use the deepspeed builtin CLI command (see here).

Is the usage of deepspeed.launcher.launch deprecated or a bad practice?

Please, consider that we are using SLURM.

Many thanks!