Compiling Learning Problems into Adaptation Programs for Language Models

2026-09-29

AI Summary

This study addresses the limitation that model adaptation relies on fixed strategies and resists dynamic optimization. To overcome this, we propose an adaptation compilation framework that reformulates the choice of adaptation method, placement, and intensity as a joint prediction-and-decision problem. The core innovation lies in constructing a counterfactual response surface that learns from historical experience to predict the performance of candidate programs. By integrating multi-objective consequence prediction with prior-guided algorithms, the framework selects optimal adaptation schemes tailored to downstream requirements without repeated search. Experiments on Llama-3.1 and Gemma-2 demonstrate that the programs selected by our framework achieve performance comparable to exhaustive search while significantly outperforming default strategies, thereby validating the amortizability of adaptation search.

Abstract

Model adaptation is typically governed by a fixed recipe, even though different update programs can produce substantially different behavioral outcomes. We introduce adaptation compilation, which reframes where, how, and to what extent a model should adapt as a joint prediction and decision problem. Rather than searching over candidate programs anew for each learning episode, a compiler learns from prior adaptations to predict a vector-valued counterfactual response surface over candidate programs---their expected effects on acquisition, transfer, boundedness, and preservation---and selects a program before adaptation begins. Because this predicted geometry captures multiple behavioral consequences rather than a single winner or scalar score, it can be reused under different downstream priorities without retraining. Across five learning types, preferred programs vary meaningfully across episodes, and this variation is predictable from pre-adaptation information. On Llama-3.1-8B, compiler-selected programs approach exhaustive search while outperforming global and objective-specific defaults. Replication on Gemma-2-9B preserves program heterogeneity and selection headroom, but shows that exploiting this headroom requires accounting for uncertainty when departing from strong defaults. Together, these results show that adaptation search can be amortized across related learning problems, turning prior adaptation experience into a basis for deciding how future learning should occur.

Problem

• Model Adaptation
• Language Models
• Adaptation Compilation
• Program Selection
• Amortized Search

Innovation

• Adaptation Compilation
• Counterfactual Response Surface
• Amortized Search
• Model Adaptation
• Language Models