SEP 4, 2026 · PREPRINT
Same Request, Different Answer: Quantization Amplifies Cache-Induced Divergence in LLM Serving
arXiv
A single-site, uncontrolled empirical measurement of nondeterminism in LLM serving systems under specific configurations; demonstrates a real phenomenon but lacks comparative intervention, power calculation, or generalization across production workloads.
PreliminaryAditi Patodiya
Reported
Trajectory divergence at 16-bit p…36.2 percent
Trajectory divergence at 4-bit qu…75.0 percent
Bit-identical reproducibility wit…0 of 800 episodes