Towards Scalable RLVR: Multimodal Instruction Following Data Synthesis and Distillation

Researchers introduced MIFS (Multimodal Instruction Following Synthesis), a pipeline that generates RL-ready multimodal data for training generalist agents, producing a dataset of 90k samples across 8 constraint categories and 14 task domains. MIFS-trained multimodal large language models (MLLMs) achieved an average improvement of 8.13% on four multimodal instruction following (MMIF) benchmarks and 3x faster training convergence versus using raw data, according to the arXiv:2609.16059v1 paper. The approach combines a generative constraint protocol, a learnability-aware distillation mechanism that filters data based on RL training dynamics, and a code-based verifier for reward signals, mitigating the generalization trade-offs typical of Supervised Fine-Tuning (SFT).

arXiv:2609.16059v1 Announce Type: new Abstract: Multimodal instruction following MMIF is crucial for building generalist agents. However, current training paradigms rely heavily on Supervised Fine-Tuning SFT , which often leads to surface-level pattern matching and degrades general capabilities. While Reinforcement Learning with Verifiable Rewards RLVR offers a promising alternative, its scalability in MMIF is severely bottlenecked by the scarcity of high-quality, RL-ready multimodal data. To bridge this gap, we present MIFS \textbf{M}ultimodal \textbf{I}nstruction \textbf{F}ollowing \textbf{S}ynthesis , a systematic pipeline designed to generate RL-ready multimodal data. Specifically, MIFS introduces a generative constraint protocol to synthesize diverse raw samples, followed by a learnability-aware distillation mechanism that filters data based on RL training dynamics to ensure stable policy optimization. Furthermore, a code-based verifier provides high-precision reward signals for policy learning. The resulting dataset comprises 90k samples across 8 constraint categories and 14 task domains. Empirical evaluations demonstrate that MIFS-trained MLLMs achieve an average improvement of 8.13\% on four MMIF benchmarks and a 3$\times$ faster training convergence compared to using raw data. Crucially, our approach mitigates the generalization trade-offs typical of SFT, preserving core visual capabilities while significantly boosting instruction-following precision.