Namely AMD would have to publish the hardware documentation (for instance command packets) of anything specific to inference (unless, this is just their GPU doi

Namely AMD would have to publish the hardware documentation (for instance command packets) of anything specific to inference (unless, this is just their GPU doing only compute).