Why LLM decode is memory-bound, not compute-bound

(github.com) by harshuljain13 | May 28, 2026 | 0 comments on HN