paper
Follow the Latent Roadmap: Navigating Revocable Decoding for Diffusion LLMs with Anchor Tokens
paperactiveprovisional
follow-the-latent-roadmap-navigating-revocable-decoding-for-diffusion-llms-with-anchor-tokens-5a71a203·1 events·first seen 25h agoAliases: Follow the Latent Roadmap: Navigating Revocable Decoding for Diffusion LLMs with Anchor Tokens
Co-occurring entities
More like this (12)
Agentic Chain-of-Thought Steering for Efficient and Controllable LLM ReasoningSelf-Augmenting Retrieval for Diffusion Language ModelsWhich Models Are Our Models Built On? Auditing Invisible Dependencies in Modern LLMsContinual LLM Upcycling: A Predictor-Gated Bank-Wise Sparsity Training Recipe for Dense-to-Sparse LLMsLearning from the Self-future: On-policy Self-distillation for dLLMsBackdoor Unlearning Generalization: A Path Toward the Removal of Unknown Triggers in LLMscode synthesis LLMsA sleep-like consolidation mechanism for LLMsExpRL: Exploratory RL for LLM Mid-TrainingJanus: A Benchmark for Goal-Conditioned Information Distortion in LLMsDiffusion Language ModelsVIA-SD: Verification via Intra-Model Routing for Speculative Decoding
Recent events (1)
ASRD: Training-free anchor-guided revocable decoding for diffusion LLMs improves accuracy and throughput
A new arXiv preprint introduces ASRD (Anchor Supervised Revocable Decoding), a training-free framework for improving decoding quality in diffusion large language models. The method addresses error propagation and local error reinforcement in revocable decoding by separating trusted 'anchor tokens' (identified via temporal consistency) from uncertain candidates, then applying anchor-guided generation and anchor-perturbed verification. Experiments on math and coding benchmarks show up to 6.4% accuracy improvement and 7.2× inference throughput gains over remasking baselines.