mlsanityData to 14 Oct 2026

Stateful Worlds, Stateless Elasticity: Exact-State Serving for Interactive World Models

arXiv 2607.10389 ↗First posted 11 Jul 2026cs.DC cs.LGcited by 0

Not on the map. Why?

Abstract (arXiv)

A persistent interactive world model keeps its running state resident on the GPU that serves it: a multi-gigabyte attention cache, almost all of it rewritten at every generation step. That state cannot be recomputed in interactive time or approximated without changing the world, so a live session pins its device. The pin is a scheduling problem. WorldMove moves a live session under one guarantee: the destination is bit-identical to the source, or nothing is installed. It relocates the cache in 18.8 ms same-node, 101x faster than save/load. It holds a checksum-verified 92.1-94.8 Gb/s on a 100 Gb fabric. At that rate the cache fits inside one interactive block. Migrating an actively generating session, it converges at a block boundary and the destination continues the world bit for bit. An admissibility condition decides each move. The move must complete inside the readout horizon, over bandwidth that covers the state plus its dirty rate. Lifted to a fleet schedulability test, it governed a consolidation loop that executed 48 of 48 migrations bit-identical across two providers. Two constraints are structural. Bit-exactness survives only inside a controlled configuration of one GPU architecture, so moving the state is the only way to preserve it exactly in interactive time. Verification cannot hide inside the wire on this fabric. Receive-path checksums stall the transport at protocol timescales under fan-in, and unscheduled incast silently collapses a receiver while every delivered byte stays correct. An incast-aware admission controller holds zero misses to 1.4x offered load and sheds overload as rejects. A lossless GPU codec widens the admission gate to fabrics raw motion cannot use. We exercise the serving loop and the mover separately, each end to end. Their composition on one fabric is unbuilt. Exact-state elasticity is a joint scheduling problem over transport and verification.

Key facts. No citing papers in the map yet (data through 14 Oct 2026). Builds on 27 of its 74 references that are in the map. First posted on arXiv 11 Jul 2026.

Builds on

27 of 74 references.

References · 27

  1. 2023Efficient Memory Management for Large Language Model Serving with PagedAttentioncited by 5,944
  2. 2023SGLang: Efficient Execution of Structured Language Model Programscited by 1,141
  3. 2024DistServe: Disaggregating Prefill and Decoding for Goodput-optimized Large Language Model Servingcited by 617
  4. 2024KVQuant: Towards 10 Million Context Length LLM Inference with KV Cache Quantizationcited by 591
  5. 2023Splitwise: Efficient generative LLM inference using phase splittingcited by 558
  6. 2024Taming Throughput-Latency Tradeoff in LLM Inference with Sarathi-Servecited by 526
  7. 2025LongLive: Real-time Interactive Long Video Generationcited by 265
  8. 2023CacheGen: KV Cache Compression and Streaming for Fast Large Language Model Servingcited by 203
  9. 2024Llumnix: Dynamic Scheduling for Large Language Model Servingcited by 174
  10. 2024BurstGPT: A Real-world Workload Dataset to Optimize LLM Serving Systemscited by 87
  11. 2024xDiT: an Inference Engine for Diffusion Transformers (DiTs) with Massive Parallelismcited by 56
  12. 2025StreamDiffusionV2: A Streaming System for Dynamic and Interactive Video Generationcited by 30
15 more
  1. 2021Silent Data Corruptions at Scalecited by 23
  2. 2026WorldKV: Efficient World Memory with World Retrieval and Compressioncited by 15
  3. 2018Revisiting Network Support for RDMAcited by 8
  4. 2026X-Cache: Cross-Chunk Block Caching for Few-Step Autoregressive World Models Inferencecited by 7
  5. 2026Execution-State Capsules: Graph-Bound Execution-State Checkpoint and Restore for Low-Latency, Small-Batch, On-Device Physical-AI Servingcited by 7
  6. 2025CRIUgpu: Transparent Checkpointing of GPU-Accelerated Workloadscited by 5
  7. 2026GENSERVE: Efficient Co-Serving of Heterogeneous Diffusion Model Workloadscited by 4
  8. 2026Bit-Exact AI Inference Verification Without Performance Tradeoffscited by 4
  9. 2026FlowPrefill: Decoupling Preemption from Prefill Scheduling Granularity to Mitigate Head-of-Line Blocking in LLM Servingcited by 3
  10. 2026WorldCache: Content-Aware Caching for Accelerated Video World Modelscited by 3
  11. 2026TurboServe: Serving Streaming Video Generation Efficiently and Economicallycited by 3
  12. 2026Demystifying Numerical Instability in LLM Inference: Achieving Reproducible Inference for Mission-Critical Tasks with HEALcited by 2
  13. 2018Fast Integrity Verification for High-Speed File Transferscited by 1
  14. 2026Adaptive Resource Management and Quality Control for Streaming Video Generationcited by 1
  15. 2026ShuntServe: Cost-Efficient LLM Serving on Heterogeneous Spot GPU Clusterscited by 1

Built on by

Not cited yet.

Similar, not cited

Close in wording, no citation either way

  1. 2026Persistent Computational State: A Session-Centric Runtime for Generative World Models
  2. 2026Programmable World Model
  3. 2026TRW: TRACE-RealWorld---An Auditable Consistency Contract for World Models as Materialized Views
  4. 2026Metronome: Bound the Cache, Keep the Beat for Real-Time Interaction Model Serving
  5. 2026Bounded-State Restoration: Decoupling Local Restore Capacity from External LLM State
  6. 2026Current World Models Lack a Persistent State Core
  7. 2026Beyond Pixel Histories: World Models with Persistent 3D State
  8. 2026ScratchWorld: Evaluating If World Models Compute Executable Consequences