Fix mem_per_cpu in SlurmIO not being passed as snake_case

[janosh]

passing memory_per_thread currently crashes job submissions with ValueError

resources = QResources(
   # ...
   memory_per_thread=10_000,
)

reason being that it's mapped to mem-per-gpu by _qresources_mapping

qtoolkit/src/qtoolkit/io/slurm.py

Line 558 in 346c0d6

"memory_per_thread": "mem-per-cpu",

but

qtoolkit/src/qtoolkit/io/slurm.py

Line 157 in 346c0d6

#SBATCH --mem-per-cpu=$${mem_per_cpu}

expects it as snake case

  File                                                                       
"/lambdafs/janosh/.venv/p312/lib/python3.12/site-packages/jobflow_remote/jo… 
line 3334, in lock_job_for_update                                            
    yield lock                                                               
  File                                                                       
"/lambdafs/janosh/.venv/p312/lib/python3.12/site-packages/jobflow_remote/jo… 
line 538, in advance_state                                                   
    states_methods[state](lock)                                              
  File                                                                       
"/lambdafs/janosh/.venv/p312/lib/python3.12/site-packages/jobflow_remote/jo… 
line 685, in submit                                                          
    submit_result = queue_manager.submit(                                    
                    ^^^^^^^^^^^^^^^^^^^^^                                    
  File                                                                       
"/lambdafs/janosh/.venv/p312/lib/python3.12/site-packages/jobflow_remote/re… 
line 155, in submit                                                          
    script_fpath = self.write_submission_script(                             
                   ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^                             
  File                                                                       
"/lambdafs/janosh/.venv/p312/lib/python3.12/site-packages/jobflow_remote/re… 
line 186, in write_submission_script                                         
    script_str = self.get_submission_script(                                 
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^                                 
  File                                                                       
"/lambdafs/janosh/.venv/p312/lib/python3.12/site-packages/jobflow_remote/re… 
line 104, in get_submission_script                                           
    return self.scheduler_io.get_submission_script(commands_list, options)   
           ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^   
  File                                                                       
"/lambdafs/janosh/.venv/p312/lib/python3.12/site-packages/qtoolkit/io/base.… 
line 57, in get_submission_script                                            
    if header := self.generate_header(options):                              
                 ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^                               
  File                                                                       
"/lambdafs/janosh/.venv/p312/lib/python3.12/site-packages/qtoolkit/io/base.… 
line 88, in generate_header                                                  
    raise ValueError(msg)                                                    
ValueError: The following keys are not present in the template: mem-per-cpu  
                                                                             
}

[janosh]

a thing to note here is that this wasn't caught by tests because it seems the template insertion isn't run by any test but probably should be?

[ml-evs]

a thing to note here is that this wasn't caught by tests because it seems the template insertion isn't run by any test but probably should be?

Yeah, I was looking into writing some tests for this as part of my review, but let's not hold this up. Thanks for reporting and fixing @janosh!