Taming the Memory Footprint Crisis: System Design for Production Diffusion LLM Serving

TL;DR

This paper introduces dLLM-Serve, a comprehensive system for efficient diffusion LLM serving that addresses memory and performance challenges, achieving significant throughput and latency improvements on diverse hardware.

Contribution

We present dLLM-Serve, a holistic serving framework with novel techniques to optimize memory, scheduling, and sparsity for diffusion LLMs in production environments.

Findings

Up to 1.81× throughput improvement on RTX 4090

Nearly 4× reduction in tail latency under contention

First scalable blueprint for diffusion LLM inference

Abstract

Diffusion Large Language Models (dLLMs) have emerged as a promising alternative to Autoregressive Models (ARMs), utilizing parallel decoding to overcome sequential bottlenecks. However, existing research focuses primarily on kernel-level optimizations, lacking a holistic serving framework that addresses the unique memory dynamics of diffusion processes in production. We identify a critical "memory footprint crisis" specific to dLLMs, driven by monolithic logit tensors and the severe resource oscillation between compute-bound "Refresh" phases and bandwidth-bound "Reuse" phases. To bridge this gap, we present dLLM-Serve, an efficient dLLM serving system that co-optimizes memory footprint, computational scheduling, and generation quality. dLLM-Serve introduces Logit-Aware Activation Budgeting to decompose transient tensor peaks, a Phase-Multiplexed Scheduler to interleave heterogeneous request phases, and Head-Centric Sparse Attention to decouple logical sparsity from physical storage. We evaluate dLLM-Serve on diverse workloads (LiveBench, Burst, OSC) and GPUs (RTX 4090, L40S). Relative to the state-of-the-art baseline, dLLM-Serve improves throughput by 1.61$\times$-1.81$\times$ on the consumer-grade RTX 4090 and 1.60$\times$-1.74$\times$ on the server-grade NVIDIA L40S, while reducing tail latency by nearly 4$\times$ under heavy contention. dLLM-Serve establishes the first blueprint for scalable dLLM inference, converting theoretical algorithmic sparsity into tangible wall-clock acceleration across heterogeneous hardware. The code is available at https://github.com/chosen-ox/dLLM-Serve.