Mask-Guided KV Cache Eviction in Block Diffusion Language Models

arXiv:2610.06996v1 Announce Type: cross Abstract: Block diffusion language models keep a large key-value (KV) cache throughout generation and attend to it at every denoising step, limiting both memory capacity and generation speed. Reducing these costs requires deciding which past tokens to use for denoising the current block (selection) and which to keep in memory for future blocks (eviction). We propose MaskAhead, a training-free method that solves both tasks with a single mask-query-based ranking mechanism. Current-block masks guide selection, while probes of upcoming masked blocks guide eviction. Both rank KV entries by their estimated contribution to the attention output. Our quantized variant, Q-MaskAhead, computes selection and attention directly from low-bit KV, largely preserving the selected entries. Experiments on Fast-dLLM-v2, DreamReasoner, and LLaDA2.0-mini cover long-generation reasoning, long-prompt question answering, and needle-in-a-haystack retrieval. On long-prompt QA, MaskAhead reduces KV memory by 9.5imes on average with a 1.2-point mean F1 loss relative to dense inference. Q-MaskAhead increases the reduction to 20.1imes with a 2.3-point mean F1 loss. In a batch-32 systems profile, MaskAhead achieves 1.23imes end-to-end and 1.68imes decode-stage speedups over dense inference.

Related

• TaskPress: Query-Agnostic KV Cache Compression via Task-Guided Pruning
• LKV: End-to-End Learning of Head-wise Budgets and Token Selection for LLM KV Cache Eviction
• DCO: Dynamic Cache Orchestration for LLM Accelerators through Predictive Management

Source: arXiv cs.AI | 2026-10-07