Same Request, Different Answer: Quantization Amplifies Cache-Induced Divergence in LLM Serving

Abstract:Prefix caching, in which a serving engine reuses the key and value tensors of a shared prompt prefix across requests, is enabled by default in the major open-source stacks and treated as a transparent optimization. We measure what it costs in reproducibility, and find that the cost rises sharply with weight quantization. Holding the model, decoding parameters, seed, and request order fixed, and issuing every request serially at batch size one, we ran an eighty-episode multi-turn agentic tool-use workload with caching enabled and disabled across two engines and four weight formats. Enabling the cache changed the agent's trajectory on 36.2 percent of episodes at 16-bit precision and on 75.0 percent at four-bit, a gradient that survives re-measurement under a controlled cache configuration. With caching disabled, repeated execution was bit-identical in every configuration, 0 of 800 episodes, which bounds other sources of nondeterminism at 0.5 percent. Repeated cache-enabled runs did diverge, and three experiments locate the cause: a single server-level prompt-cache setting moves run-to-run divergence by 37.5 percentage points, execution order acts only while that setting is active, and restoring cache state makes the cached and recompute paths each reproduce on 40 of 40 items while still differing from each other on 14. Cached serving is deterministic given cache state, and irreproducible in practice because that state is absent from the request and never reset by default. A single-turn bridge shows the divergence reaching task outcomes without shifting aggregate accuracy. We release the harness, logs, and analysis pipeline.

Software Engineering (cs.SE); Distributed, Parallel, and Cluster Computing (cs.DC); Machine Learning (cs.LG)

arXiv:2609.04748 [cs.SE]

(or arXiv:2609.04748v1 [cs.SE] for this version)

https://doi.org/10.48550/arXiv.2609.04748 Focus to learn more arXiv-issued DOI via DataCite (pending registration)

arXivLabs is a framework that allows collaborators to develop and share new arXiv features directly on our website.

Both individuals and organizations that work with arXivLabs have embraced and accepted our values of openness, community, excellence, and user data privacy. arXiv is committed to these values and only works with partners that adhere to them.

Have an idea for a project that will add value for arXiv's community? Learn more about arXivLabs.