Abstract:A modern post-training pipeline often writes one symbol for its policy, pi_theta, while evaluating it through two different programs: a training kernel optimized for autograd and an inference kernel optimized for low-precision, fused, dynamically batched serving. In finite precision, these kernels can induce different distributions at identical weights, with the gap concentrated on slices that aggregate benchmarks under-represent. This paper proposes kernel contracts: a contract-first framework for specifying acceptable divergence between K_train and K_inf. A contract C = (N, S, R, O, Pi) combines numerical, statistical, runtime, and observability clauses with an escalation policy from violations to routing actions. We derive a chain of bounds from logit drift to total-variation distance to bounded reward drift, and specialize it to RL post-training, where per-token importance-ratio drift yields a bound on policy-gradient bias under explicit support and norm assumptions. We also describe a four-stage promotion pipeline, online routing loop, and minimal YAML DSL for contract artifacts. This is a framework and vocabulary paper; we do not report production-scale empirical validation.
From: Bruce Xu [view email]
[v1]
Tue, 26 May 2026 21:48:33 UTC (28 KB)