ParetoBandit uses contextual bandits with an online primal-dual budget pacer and geometric forgetting to enforce cost ceilings and adapt to non-stationary pricing and quality shifts in LLM serving, achieving 0.4% budget compliance and up to 0.071 quality lift on 1824 prompts.
Garivier and Moulines (Garivier & Moulines, 2011) derive γ as a function of the (unknown) number of breakpoints, which is not observable in production
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.LG 1years
2026 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
ParetoBandit: Budget-Paced Adaptive Routing for Non-Stationary LLM Serving
ParetoBandit uses contextual bandits with an online primal-dual budget pacer and geometric forgetting to enforce cost ceilings and adapt to non-stationary pricing and quality shifts in LLM serving, achieving 0.4% budget compliance and up to 0.071 quality lift on 1824 prompts.