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

2026-07-11

AI Summary

This work addresses the challenge of interactive world models being locked to specific devices due to their reliance on multi-gigabyte attention caches on GPUs, which are impractical to recompute or migrate within interactive latency constraints. To overcome this, the authors propose WorldMove, a mechanism that jointly ensures bit-exact state migration, incast-aware admission control, and lossless GPU-based encoding/decoding. For the first time, this approach achieves both state precision and elastic scheduling. Experimental results demonstrate that intra-node migration completes in just 18.8 ms—101× faster than conventional store-and-load methods—and achieves verification throughput of 92.1–94.8 Gb/s over a 100 Gb/s network. The system also successfully executed 48 zero-failure, bit-exact migrations across different cloud providers.

Abstract

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.

Problem

• interactive world models
• stateful serving
• exact-state migration
• GPU elasticity
• real-time scheduling

Innovation

• exact-state migration
• interactive world models
• GPU state transfer
• incast-aware admission control
• bit-identical elasticity