You are right, this is the memory documentation that is correct, but not precise enough, this should be something like:

• memory: Total amount of memory to be used by all workers inside a job. Used by job queuing system by default as amount of memory per job.
• job_mem: Amount of memory to request in SLURM for each job, If None, defaults to memory.

I agree with the improvements mentioned here.
Similarly, can we modify the description of "cores" and "job_cpu" parameters too?

Something like this:

• cores: Total number of CPU cores on which all the worker threads inside a job will run. The number of threads per worker process are determined using formula cores / processes. Used by job queuing system by default as amount of CPUs per job.

• job_cpu: Number of CPUs to request in SLURM for each job. This option is useful to request more CPUs than total number of worker threads, in cases when a dask task might run multi-threaded using OpenMP or other means. If None, defaults to cores.

@guillaumeeb Would like to know your thoughts on the above comment regarding description of CPU parameters.

Thanks @maneesh29s for the proposal and sorry for the delay.

Just a small suggestion:
job_cpu: Number of CPUs to request in SLURM for each job. This option might be useful to request more CPUs than total number of worker threads for some complex non Python code. If None, defaults to cores.

I agree with your suggestion.

And I am happy to raise a PR to close this issue.

Sure, please do so!