FineServe: A Fine-Grained Dataset and Characterization of Global LLM Serving Workloads

Collection of new ML/AI systems and benchmarks with multiple security-relevant contributions. Key items: FineServe — a fine-grained, in-the-wild multi-model LLM serving workload dataset and workload generator for realistic routing/scheduling/capacity planning; FraudShield AI — an LSTM+graph-topology hybrid for resilient financial-fraud detection under extreme class imbalance; OpenEvoShield — a co-evolutionary continual-defense framework for adversarial instruction injection in LLM-based multi-agent systems; a benchmark of confidential GPU inference (NVIDIA H100 under Intel TDX) quantifying 11–