Summary
Compile by training is a technique that converts natural-language specifications into reusable neural functions. At compile time, a large teacher model generates task-specific examples used to train a small adapter on a compact interpreter; the resulting function runs locally without needing the teacher at inference time, and can be stored, versioned, and composed like ordinary software. This avoids repeated costs, latency, and provider dependency associated with calling large remote models per input. On FuzzyBench-Hard, where a fast Program-as-Weights compiler produced no exact matches, compile by training achieves 83.6% semantic accuracy. Higher accuracy comes with higher compile-time cost: roughly one minute versus seconds for the fast compiler. The authors—Yuntian Deng, Pengyu Nie, and Stuart Shieber—deploy the compiler in a public interactive service and demonstrate compiled functions in a multi-site website assistant, a language-controlled 3D avatar, and a bidirectional English-Claudish translator.
论文概要
Research area: NLP
Authors: Yuntian Deng, Pengyu Nie, Stuart Shieber
Published: 2026-09-03
arXiv: 2609.04199
Summary
Many recurring text-processing functions are easy to describe in natural language but hard to implement with rules, while calling a large remote model for every input incurs repeated costs, latency, and provider dependency. The authors propose compile by training, which turns a natural-language specification into a reusable neural function.
How it works
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.
Results
On FuzzyBench-Hard—a subset where the fast Program-as-Weights compiler failed to produce any exact matches—compile by training reaches 83.6% semantic accuracy. Higher accuracy comes with higher compile-time cost: about one minute, versus seconds for the fast compiler.
Deployments
The compiler is deployed in a public interactive service, with demonstrations including:
- A multi-site website assistant
- A language-controlled 3D avatar
- A bidirectional English–Claudish translator
---
Source: arXiv:2609.04199
*Auto-collected on 2026-09-05*
This page is an English static mirror generated for search and AI citation.
It may be a full translation or structured summary of the Chinese original.
Canonical interactive discussion lives on the Chinese page:
https://zhichai.net/topic/178634488