dLLM-Cache: New Caching Framework Accelerates Diffusion-Based Language Models
Researchers have developed dLLM-Cache, a training-free caching framework designed to speed up inference in diffusion-based large language models (dLLMs), which generate text through iterative denoising rather than traditional autoregressive methods. The method exploits the observation that most tokens remain stable across denoising steps, combining prompt caching with partial response updates to reduce computational overhead. The approach achieves up to 9.1x reduction in floating-point operations while maintaining output quality, potentially making dLLMs competitive with traditional autoregressive models in terms of inference speed.
Diffusion-based large language models (dLLMs) represent an emerging alternative to traditional autoregressive models, generating text by iteratively denoising masked segments. However, they suffer from high inference latency, and existing acceleration techniques like Key-Value caching cannot be directly applied due to their bidirectional attention mechanism. The dLLM-Cache framework addresses this by leveraging a key insight: during inference, the prompt remains static while the response is partially dynamic, with most tokens staying stable between adjacent denoising steps. The method combines long-interval prompt caching with feature similarity-guided partial response updates, enabling efficient reuse of intermediate computations without training. Experiments on models including LLaDA 8B and Dream 7B demonstrate up to 9.1x FLOPs reduction on LongBench-HotpotQA benchmarks while preserving competitive output quality, bringing dLLM inference latency close to that of traditional autoregressive models in many scenarios.
What's missing
The paper does not discuss potential limitations of the approach, such as scenarios where the assumption about token stability may not hold, memory overhead of maintaining caches, or how performance scales with different model sizes beyond the two tested examples.
What different sources said
- arXiv cs.CLCenter
Diffusion Language Model Parallel Decoding via Product-of-Experts Bridge
Related
Gut Bacteria Enzyme Found to Break Down Heat-Processed Food Compounds, Producing Novel Biogenic Amines
Researchers have discovered that an enzyme in common gut bacteria can degrade N-epsilon-carboxymethyllysine (CML), a compound formed during thermal food processing, producing previously unknown biogenic amines. The enzyme, ornithine decarboxylase SpeC from enterobacteria, acts on CML and related modified lysine derivatives through a low-level 'underground' catalytic activity. This finding suggests a previously unrecognized communication axis between thermally processed dietary compounds and gut microbial physiology, with potential implications for host health.
Full-Length Gene Sequencing Reveals Two Distinct Bacterial Communities in Black-Legged Ticks Expanding Into Canada
Researchers used Oxford Nanopore full-length 16S rRNA gene sequencing to characterize the microbiome of Ixodes scapularis black-legged ticks collected in Nova Scotia, Canada, distinguishing between tick-adapted bacteria and environmentally acquired bacteria. The study comes as I. scapularis — the primary vector of Lyme disease — is rapidly expanding northward into Canada due to climate change. The findings suggest that environmentally derived bacteria in tick microbiomes are not mere contamination, which has implications for how tick microbiome data is collected and interpreted across surveillance studies.
Study Identifies Metabolic Link Between Cell Envelope Stress and Biofilm Formation in Bacteria
Researchers have discovered that the metabolite acetyl-CoA directly inhibits enzymes that degrade the bacterial signaling molecule c-di-GMP, connecting cell envelope biosynthesis stress to biofilm formation in Pseudomonas aeruginosa. The study found that sub-inhibitory concentrations of antibiotics targeting early peptidoglycan biosynthesis — but not other antibiotic classes — elevate c-di-GMP levels by reducing phosphodiesterase activity, with acetyl-CoA competing for the enzyme active site. Because the relevant enzyme domain is broadly conserved across bacterial species, this checkpoint mechanism may be widespread and could have implications for understanding antibiotic-induced biofilm responses.