[measure.py] -ms / max_sys argument is unclear

[mratsim]

The argument states in its help that it's a limit for state data

exllamav3/exllamav3/conversion/measure_model.py

Line 31 in 2a4aac0

parser.add_argument("-ms", "--max_sys", type = float, default = 8, help = "Max system memory for state data, in GB")

However, while setting to 32GiB, the process consumed about 62GiB when doing a 3-way measurements or 130GiB with a 5-way measurement.

It would be useful to either limit the max memory usage, all included, or have an heuristic for max mem usage (+max_sys per extra quantized model measured over the first 2?)