Stop the Flip-Flop: Context-Preserving Verification for Fast Revocable Diffusion Decoding

2026-02-05

AI Summary

This work addresses the issue of flip-flop oscillations in aggressive parallel decoding within diffusion-based language models, which often lead to context dilution and ineffective revisions, thereby degrading inference efficiency. To mitigate this, the authors propose COVER, a method that constructs dual attention views via KV cache overwriting to enable leave-one-out validation and stable draft generation within a single forward pass. This approach effectively prevents self-leakage while preserving contextual information. COVER further incorporates a context-preserving verification mechanism, stability-aware seed selection, and dynamic adjustment of the number of verified tokens, collectively suppressing flip-flop oscillations. Experimental results demonstrate that COVER significantly reduces redundant revisions, accelerates decoding speed, and maintains high generation quality.

Technology Category

• Natural Language Processing: (Large) Language Models
• Machine Learning: Large Multimodal Models (LMMs)
• Computer Vision: Diffusion Models for Vision

Application Category

• Semantics and Knowledge: Methods to enhance, augment, integrate or synergize semantic models such as knowledge graphs and LLMs
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• User Modeling, Personalization and Recommendation: Attacks and countermeasures in recommendation systems

Abstract

Parallel diffusion decoding can accelerate diffusion language model inference by unmasking multiple tokens per step, but aggressive parallelism often harms quality. Revocable decoding mitigates this by rechecking earlier tokens, yet we observe that existing verification schemes frequently trigger flip-flop oscillations, where tokens are remasked and later restored unchanged. This behaviour slows inference in two ways: remasking verified positions weakens the conditioning context for parallel drafting, and repeated remask cycles consume the revision budget with little net progress. We propose COVER (Cache Override Verification for Efficient Revision), which performs leave-one-out verification and stable drafting within a single forward pass. COVER constructs two attention views via KV cache override: selected seeds are masked for verification, while their cached key value states are injected for all other queries to preserve contextual information, with a closed form diagonal correction preventing self leakage at the seed positions. COVER further prioritises seeds using a stability aware score that balances uncertainty, downstream influence, and cache drift, and it adapts the number of verified seeds per step. Across benchmarks, COVER markedly reduces unnecessary revisions and yields faster decoding while preserving output quality.

Problem

• revocable decoding
• flip-flop oscillations
• diffusion language models
• parallel decoding
• verification

Innovation

• revocable decoding
• diffusion language models
• KV cache override
• context preservation
• stable drafting