See parent issue for context.
As shown in the diagram above, there are some caveats to this one that might make this a bit trickier than originally thought (there may be some refactoring involved).
My initial thinking is this: for each selection combination of Model + ISL/OSL + Precision, you can "calculate" the set of all possible calculations of GPU (hw field which is truncated as described here) + Framework (framework field after mapping applied, as described here + additional software stuff (such as mtp field)1. Once you have the set of all these combinations for a particular selection, you use these as selectable configurations on the legend and on the "Select a GPU for comparison" dropdown. This means that the "Select a GPU for comparison" dropdown can no longer be static and will depend on the selected Model + ISL/OSL + Precision.
See parent issue for context.
As shown in the diagram above, there are some caveats to this one that might make this a bit trickier than originally thought (there may be some refactoring involved).
My initial thinking is this: for each selection combination of Model + ISL/OSL + Precision, you can "calculate" the set of all possible calculations of GPU (
hwfield which is truncated as described here) + Framework (frameworkfield after mapping applied, as described here + additional software stuff (such asmtpfield)1. Once you have the set of all these combinations for a particular selection, you use these as selectable configurations on the legend and on the "Select a GPU for comparison" dropdown. This means that the "Select a GPU for comparison" dropdown can no longer be static and will depend on the selected Model + ISL/OSL + Precision.Footnotes
Another annoying detail, with this PR feat: multinode first class integration #251, the
mtp: on/offfield has been changed tospec-decoding: [str]. Therefore, we will have to support both themtpfield and the newerspec-decodingfield. ↩