STaR-Quant: New Quantization Method Improves Efficiency of Diffusion Language Models
Researchers have proposed STaR-Quant, a post-training quantization framework designed to reduce the memory and computational costs of diffusion large language models (DLLMs). DLLMs generate text through iterative masked denoising rather than the token-by-token approach of autoregressive models, but their size and iterative process create significant deployment overhead. The work addresses this by introducing techniques to handle activation inconsistencies across token states and error accumulation across denoising steps, achieving up to 1.69x speedup and 3.14x memory reduction.
Diffusion large language models (DLLMs) have emerged as an alternative to autoregressive language models, generating text by iteratively denoising masked tokens using bidirectional context. However, their large parameter counts and multi-step inference process impose substantial memory and compute burdens, making efficient deployment challenging. The authors of STaR-Quant identify two core obstacles to low-bit quantization of DLLMs: state-dependent activation disparity, where masked and unmasked tokens exhibit different activation distributions within a single denoising step, and temporal error accumulation, where quantization errors compound across iterative decoding steps. To address these, STaR-Quant introduces State-Guided Activation Transformation (SGAT), which routes masked and unmasked tokens into separate activation transformation spaces while sharing a single static weight-side transformation, and Temporal Attention Compensation (TAC), a lightweight block-diagonal affine mapping that corrects quantized attention representations across steps. Experiments on representative DLLMs show consistent improvements over strong post-training quantization baselines at low bit-widths. The framework achieves up to 1.69x inference speedup and 3.14x memory savings compared to FP16 deployment. The preprint was submitted to arXiv in early June 2026 and has not yet undergone formal peer review.
What's missing
The work is a preprint and has not yet been peer-reviewed. It is also unclear how STaR-Quant performs relative to quantization methods developed for autoregressive LLMs when those are adapted to DLLMs.
What different sources said
- arXiv cs.LGCenter
STaR-Quant: State-Time Consistent Post-Training Quantization for Diffusion Large Language Models
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.