Track papers across labs, topics, and trends.
Search papers, labs, and topics across Lattice.
TRACER is introduced as a system that trains lightweight ML surrogates on LLM production traces for cost-efficient classification, routing traffic to the surrogate only when its agreement with the LLM exceeds a threshold. This approach leverages readily available production logs to create a "free" training set, reducing inference costs by offloading a significant portion of traffic from the LLM. Experiments on intent classification benchmarks show TRACER achieves high surrogate coverage (83-100%) and can even fully replace the LLM teacher in some cases, while a parity gate prevents deployment when the surrogate is unreliable.
Stop paying for every LLM classification call: TRACER uses your LLM's own production logs to train free, lightweight surrogates that can handle most requests.
Every call to an LLM classification endpoint produces a labeled input-output pair already retained in production logs. These pairs constitute a free, growing training set: a lightweight surrogate trained on them can absorb a significant portion of future traffic at near-zero marginal inference cost. The open questions are when the surrogate is reliable enough to deploy, what it handles versus defers, and how that boundary evolves as data accumulates. We introduce TRACER (Trace-based Adaptive Cost-Efficient Routing), an open-source system that trains ML surrogates on an LLM's own production traces and governs deployment through a parity gate: the surrogate is activated only when its agreement with the LLM exceeds a user-specified threshold {\alpha}. To make the routing boundary transparent, TRACER generates interpretability artifacts describing which input regions the surrogate handles, where it plateaus, and why it defers. On a 77-class intent benchmark with a Sonnet 4.6 teacher, TRACER achieves 83-100% surrogate coverage depending on the quality target {\alpha}; on a 150-class benchmark, the surrogate fully replaces the teacher. On a natural language inference task, the parity gate correctly refuses deployment because the embedding representation cannot support reliable separation. The system is available as open-source software.