infrastructurefactneutralOperators currently lack tools to match specific LLMs to the most efficient GPUs without exhaustively profiling each combinationMachine Learning27 Jul 2026http://arxiv.org/abs/2607.02391v1