Masked Language Flow Models

2026-06-25

AI Summary

Existing streaming language models struggle to support multi-step reasoning due to their requirement of generating all tokens in a single pass, while masked diffusion models suffer from degraded generation quality under few-step sampling owing to independence assumptions. This work presents the first integration of masking mechanisms with continuous flow modeling, introducing a conditional generation framework based on continuous stochastic interpolation that enables lightweight adaptation from pretrained masked diffusion models. The proposed hybrid sampling strategy alternates between continuous denoising and confidence-driven discrete unmasking, facilitating efficient multi-step reasoning. Experiments demonstrate significant performance gains on GSM8K and MT-Bench, marking the first scalable application of streaming language models to complex reasoning and instruction-following tasks.

Abstract

Masked Diffusion Models (MDMs) promise fast, parallel language generation, but their reverse transition factorises across token positions -- an approximation that breaks down in the few-step sampling regime where parallel generation ought to provide the greatest efficiency gains. Flow Language Models (FLMs) sidestep this limitation by learning a continuous flow that transports noise toward clean sequences represented in Euclidean space, inducing a flow map that can be distilled for single-step generation. However, this makes complex tasks requiring multi-step reasoning problematic for FLMs, as FLMs are forced to decode every token during generation. To address this, we introduce Masked Language Flow Models (MLFMs), which incorporate masking into FLMs using a continuous stochastic interpolant to bridge partially masked and clean sequences. This design enables conditional generation via continuous flows and allows pretrained MDMs to be converted into MLFMs through a simple, lightweight adaptation. Leveraging this flexibility, we propose a novel sampler that alternates continuous denoising with the discrete unmasking of confident tokens to better support multi-step reasoning. We evaluate our approach on GSM8K and MT-Bench and find, for the first time, that flow-based language models can be scaled to solve downstream reasoning and instruction-following tasks.

Problem

• flow language models
• masked diffusion models
• multi-step reasoning
• conditional generation
• language generation

Innovation

• Masked Language Flow Models
• continuous stochastic interpolant
• flow-based language models
• multi-step reasoning
• conditional generation