BlockLock：固定批几何实现可复现 LLM 推理
原标题：BlockLock: Reproducible LLM Inference with Fixed Batch Geometry
AI 摘要
论文提出 BlockLock，一种让 LLM 推理在批处理中保持可复现的方法：通过固定受保护请求所见的张量几何形状（BlockLock-Rows 固定行宽并保留批量解码，BlockLock-Pack 采用终端锚定打包并隔离 dummy 段），使目标请求的输出在其他请求变化时保持逐位一致。在 Qwen3、Llama-3.1、Gemma-3 等五个稠密模型和 A6000/A100/H100 三种单卡部署上，两种模式均保持目标序列与 Top20 logit 位一致；相比 vLLM Batch-Invariant，BlockLock-Rows 延迟降低 53.5–55.0%，吞吐提升 115.2–122.2%，且无需自定义 kernel。
正文节选
marginparsep has been altered. topmargin has been altered. marginparwidth has been altered. marginparpush has been altered. The page layout violates the ICML style. Please do not change the page layout, or include packages like geometry, savetrees, or fullpage, which change it for you. We’re not able to reliably undo arbitrary changes to the style. Please remove the offending package(s), or layout-changing commands and try again. BlockLock: Reproducible LLM Inference with Fixed Batch Geometry Xi