UltraIF: Advancing Instruction Following from the Wild

TL;DR

UltraIF is a scalable method that decomposes complex instructions into simpler components, enabling open-source LLMs like LLaMA-3.1-8B to match the instruction-following performance of proprietary models without additional benchmark data.

Contribution

We introduce UltraIF, a novel approach that decomposes and synthesizes complex instructions, significantly improving open-source LLMs' instruction-following capabilities without relying on benchmark data.

Findings

LLaMA-3.1-8B-Base matches instruct models on 5 benchmarks

UltraIF improves LLaMA-3.1-8B-Base's instruction-following performance

Self-alignment with UltraIF further enhances LLaMA-3.1-8B-Instruct

Abstract

Instruction-following made modern large language models (LLMs) helpful assistants. However, the key to taming LLMs on complex instructions remains mysterious, for that there are huge gaps between models trained by open-source community and those trained by leading companies. To bridge the gap, we propose a simple and scalable approach UltraIF for building LLMs that can follow complex instructions with open-source data. UltraIF first decomposes real-world user prompts into simpler queries, constraints, and corresponding evaluation questions for the constraints. Then, we train an UltraComposer to compose constraint-associated prompts with evaluation questions. This prompt composer allows us to synthesize complicated instructions as well as filter responses with evaluation questions. In our experiment, for the first time, we successfully align LLaMA-3.1-8B-Base to catch up with its instruct version on 5 instruction-following benchmarks without any benchmark information, using only 8B model as response generator and evaluator. The aligned model also achieved competitive scores on other benchmarks. Moreover, we also show that UltraIF could further improve LLaMA-3.1-8B-Instruct through self-alignment, motivating broader use cases for the method. Our code is available at https://github.com/kkk-an/UltraIF.