Compile by Training: Turning Natural-Language Specifications into Local Neural Functions
Yuntian Deng, Pengyu Nie, Stuart Shieber
This paper introduces a method to compile neural functions from natural-language specifications, allowing for faster and more reliable execution without relying on remote models. Practitioners may care about this approach for building efficient and flexible AI systems that can perform complex tasks without the need for expensive model calls.
Abstract
Many recurring text functions are easy to describe but difficult to implement with rules, while calling a large remote model for every input introduces repeated cost, latency, and dependency on a provider. We present compile by training, which turns a natural-language specification into a reusable neural function. At compile time, teacher models generate task-specific examples that are used to train a small adapter for a compact interpreter. The resulting function runs without the teachers and can be stored, versioned, and composed like ordinary software. On FuzzyBench-Hard, a subset on which the Program-as-Weights fast compiler produced no exact matches, compile by training reaches 83.6% semantic accuracy. This higher accuracy comes with a higher compile-time cost: roughly a minute rather than seconds for the fast compiler. We deploy the compiler in a public interactive service and demonstrate compiled functions in a multi-site website helper, a language-controlled 3D avatar, and a bidirectional English-Claudish translator.