--chunks_per_runner value explanation

[sirselim]

I was just reviewing some of the code and thought that I should note that the --chunks_per_runner value of 412 has been optimised for a system with 2x A5000 GPUs. I recommend that people optimise this parameter for their particular GPUs. Code and documentation to do so can be found here.

In general this hardcoded value (412) will work for most GPUs with 24GB or RAM or more, but there is potential to get faster basecalling rates by increasing this value.