Search papers, labs, and topics across Lattice.
This paper addresses the challenge of effectively reusing past update evidence in the post-training of large language models (LLMs) by introducing a novel method called Boundary-Calibrated Intervention Transfer (BCIT). BCIT evaluates the applicability of past experiences to current contexts, ensuring that only relevant updates are authorized, which prevents degradation of the training trajectory and optimizes resource use. The method was tested on a 4B model across diverse tasks, demonstrating that BCIT leads to fewer harmful updates and superior final-model quality compared to traditional approaches.
Conditional experience transfer can significantly enhance LLM post-training by preventing harmful updates and improving model quality.
Large language models offer broad capabilities, but adapting them to evolving domains, tools, and requirements often entails repeated post-training. Autonomous systems automate parts of this process by proposing updates, training candidates, and using evaluation feedback to select subsequent proposals. As evidence accumulates, a central problem emerges: which past update evidence remains actionable after subsequent training has changed the parent model? An update's effect depends on its parent, data, and training stage. Treating past success as context-free permission can waste compute. If the resulting child is promoted, it can also degrade the subsequent training trajectory. We formulate this problem as conditional experience transfer and introduce Boundary-Calibrated Intervention Transfer (BCIT), a method that authorizes experience reuse before weight-changing training. BCIT binds an observed effect to its source context, checks applicability conditions, vetoes candidates with named hard conflicts, and obtains current-state evidence through a bounded training trial when needed. Fully trained candidates still face a shared adoption rule, and only observed events extend memory. On one 4B model adapted across finance reasoning, text-to-SQL, and function calling, candidate updates exhibit heterogeneous target and retention effects across the evaluated contexts. Under matched candidates, evidence, and compute, BCIT authorizes fewer harmful updates and attains higher equal-budget final-model quality than the evaluated alternatives. These results support treating experience authorization as a distinct problem in autonomous post-training.
