🤖 AI Summary
Existing generative world models exhibit poor performance in backtracking simulations—a limitation often mistakenly attributed to model capacity but actually stemming from deficiencies in runtime state management. This work proposes a session-centric runtime architecture that explicitly distinguishes between recomputable and non-recomputable states, introducing the concept of the minimal non-recomputable state (PCS) and enabling its efficient snapshotting and restoration. By capturing observations, random number generator states, memory banks, and sliding KV contexts, and employing a relevance-based rather than recency-based memory eviction policy, the system achieves state recovery in just 0.012 milliseconds—orders of magnitude faster than the 1.85 seconds required per generation step. The approach supports up to 1,024 concurrent sessions while ensuring exact backtracking fidelity and significantly enhancing scalability.
📝 Abstract
Generative world models are increasingly driven as simulators: a planner forks a state, rolls out futures, backtracks, and returns to a visited viewpoint. Recent benchmarks establish that current video world models fail this usage, and attribute it to the model, prescribing new architectures and training objectives. We show this attribution is incomplete, and for an important class of models simply wrong. Snapshotting the state the runtime already holds -- an observation plus RNG state, a memory bank, or a windowed KV context, by architecture -- and restoring it after a genuine excursion reproduces the never-left continuation byte-identically on all three; corrupting only the RNG degrades it. The capability was never missing: request-centric serving discarded it, inheriting from language-model serving the assumption that runtime state is recomputable -- but world-model state carries a non-recomputable kernel. We define Persistent Computational State (PCS), the minimal non-recomputable state that must survive across requests, show it can be discovered by measurement, and build a session-centric runtime over it. Checkpoint and restore cost 0.012 ms against a 1.85 s generation step; resident sessions become host- rather than device-bounded (measured to 1,024); and world memory must be evicted by relevance to the return, not recency -- the inverse of LLM practice.