Spark Policy Toolkit: Semantic Contracts and Scalable Execution for Policy Learning in Spark

This feed contains recent systems and ML infrastructure research across ten arXiv papers: (1) Spark Policy Toolkit — introduces Spark-native, semantics-governed primitives (mapInPandas/mapInArrow and collect-less split search) and a fixed-input contract to enable scalable, semantics-preserving policy learning at Spark scale with high throughput on Databricks. (2) CacheFlow — presents a 3D-parallelism abstraction (tokens, layers, GPUs) and a batch-aware two-pointer scheduler to reduce KV-cache restoration TTFT for long-context LLM serving by 10–62%. (3) Adaptive Management of Microservices — a