mirror of
https://github.com/hanzoai/papers.git
synced 2026-10-22 17:20:17 +00:00
9-page paper (mechanism/brain split, memory-aware placement, bilinear utility + LinUCB, SLO objective quality - lambda*cost - mu*latency). Itemized cost model beyond API tokens: honest local-compute electricity figure (~$0.0001/1k tok, not free), cloud-GPU-rental-avoided, run-for-pay fleet angle, seat consolidation, overage avoidance, 1% routing fee, and a 20-person-team TCO table. Builds clean via pdflatex+bibtex.