I think the current renames are accurate, but it's hard to be certain. I never found good documentation which clearly maps low-level names to human-readable names.
Among other things, we should ensure that we are not renaming any low-level subsets to their human-readable supersets. For example, CIK-KABINI is part of AMD's Radeon HD 8000 series, but it's not the only chipset in that series, so it would seem that it would be inaccurate to rename CIK-KABINI to Radeon HD 8000. (We may want to leave the name CIK-KABINI as-is for that reason.) There may be other cases where we should leave the low-level names as-is.
The hardware.json file renames some GPU models from their low-level names to more human-readable names.
https://github.com/mozilla/ensemble-transposer/blob/22e787f559ec56ce2185a7643d8aabbba208172f/config/datasets/hardware.json#L43-L192
I think the current renames are accurate, but it's hard to be certain. I never found good documentation which clearly maps low-level names to human-readable names.
Among other things, we should ensure that we are not renaming any low-level subsets to their human-readable supersets. For example, CIK-KABINI is part of AMD's Radeon HD 8000 series, but it's not the only chipset in that series, so it would seem that it would be inaccurate to rename CIK-KABINI to Radeon HD 8000. (We may want to leave the name CIK-KABINI as-is for that reason.) There may be other cases where we should leave the low-level names as-is.