mlsanityData to 14 Oct 2026

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

arXiv 2512.17077 ↗First posted 18 Dec 2025cs.DCcited by 5

On the map: Efficient LLM inference and serving → Serving systems and request scheduling

Abstract (arXiv)

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×-1.81× on the consumer-grade RTX 4090 and 1.60×-1.74× on the server-grade NVIDIA L40S, while reducing tail latency by nearly 4× 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.

Key facts. Cited by 5 papers in the map (data through 14 Oct 2026). Placed in Serving systems and request scheduling, one of 12 directions in Efficient LLM inference and serving. Builds on 8 of its 20 references that are in the map. First posted on arXiv 18 Dec 2025.

Builds on

8 of 20 references.

References · 8

  1. 2023FlashAttention-2: Faster Attention with Better Parallelism and Work Partitioningcited by 2,598
  2. 2023FlexGen: High-Throughput Generative Inference of Large Language Models with a Single GPUcited by 575
  3. 2025Fast-dLLM: Training-free Acceleration of Diffusion LLM by Enabling KV Cache and Parallel Decodingcited by 420
  4. 2025Accelerating Diffusion LLMs via Adaptive Parallel Decodingcited by 112
  5. 2025Sparse-dLLM: Accelerating Diffusion LLMs with Dynamic Cache Evictioncited by 65
  6. 2025AdaBlock-dLLM: Semantic-Aware Diffusion LLM Inference via Adaptive Block Sizecited by 31
  7. 2025SparseD: Sparse Attention for Diffusion Language Modelscited by 17
  8. 2025Mask Tokens as Prophet: Fine-Grained Cache Eviction for Efficient dLLM Inferencecited by 17

Built on by

Cited by year (2026 to 14 Oct)

Citing papers per year: 2026: 5.

5 papers.

  1. 2026TIDE: Efficient and Lossless MoE Diffusion LLM Inference with I/O-aware Expert Offloadcited by 2
  2. 2026Sangam: Efficiently Serving Diffusion LLMs with the AR Stackcited by 2
  3. 2026BlockServe: Block-Grained Continuous Batching for High-Throughput Diffusion LLM Servingcited by 0
  4. 2026Serving Masked Diffusion LLMs: Characterization and Design Principles from Real Hardwarecited by 0
  5. 2026Affix Cache for Diffusion Large Language Modelscited by 0

Similar, not cited

Close in wording, no citation either way

  1. 2026Accelerating Masked Diffusion Large Language Models: A Survey of Efficient Inference Techniques
  2. 2024Tackling the Dynamicity in a Production LLM Serving System with SOTA Optimizations via Hybrid Prefill/Decode/Verify Scheduling on Efficient Meta-kernels
  3. 2026Understanding Efficiency: Quantization, Batching, and Serving Strategies in LLM Energy Use
  4. 2026Mosaic: Unlocking Long-Context Inference for Diffusion LLMs via Global Memory Planning and Dynamic Peak Taming
  5. 2026DiLaServe: High SLO Attainment Serving for Diffusion Language Models
  6. 2025SeaLLM: Service-Aware and Latency-Optimized Resource Sharing for Large Language Model Inference
  7. 2026ES-dLLM: Efficient Inference for Diffusion Large Language Models by Early-Skipping
  8. 2026SLIM: Saturation-Aware Lightweight Performance Modeling for LLM Serving