Search papers, labs, and topics across Lattice.
This paper introduces Information-Guided Frontier Decoding (IGFD), a novel decoding strategy for diffusion multimodal language models (dMLLMs) that prioritizes the commitment of semantically informative tokens over structurally easy ones. By integrating token confidence, neighborhood uncertainty, and structural commitment risk, IGFD enhances contextual support during the decoding process, leading to improved performance across various benchmarks. Experimental results demonstrate that IGFD significantly outperforms existing decoding methods without requiring additional training or resources, thus advancing the efficiency and effectiveness of dMLLMs.
Prioritizing semantic anchors over easy tokens can drastically reduce error accumulation in multimodal language models.
Decoding quality in diffusion multimodal language models (dMLLMs) depends heavily on the order in which masked tokens are committed. Existing confidence-based strategies prioritize locally easy tokens, but confidence does not necessarily reflect contextual usefulness. As a result, structurally easy tokens such as punctuation may be committed before informative semantic anchors, weakening context propagation and increasing error accumulation. We propose Information-Guided Frontier Decoding (IGFD), a training-free decoding strategy that ranks candidates using token confidence, neighborhood uncertainty, and structural commitment risk. IGFD encourages early commitment of reliable semantic anchors while delaying fragile structural tokens, improving contextual support during decoding. A dynamic candidate frontier further constrains token selection to locally expandable regions under the same decoding budget. The method requires no additional training, auxiliary models, or extra forward passes. Experiments across multimodal understanding, reasoning, grounding, and hallucination benchmarks show that IGFD consistently outperforms existing decoding strategies across the majority of benchmarks and diffusion MLLM backbones under identical decoding budgets.
