Architectural Reference and System Flow Documentation for Distributed LLM Inference. Mapping the control-plane integration between the Kubernetes Gateway API Inference Extension and the LLM-D Inference Scheduler data-plane.
Trust-Weighted, Fault-Tolerant Native Kubernetes Scheduler Plugin in Go. Combines predictive ACO optimization, gRPC sidecar, Trust Discounting (κ_i) for flapping telemetry, Go Circuit Breaker failover, and GPU QoS routing. <0.98ms P99 latency at 1,250 pods/sec. 46.3% cost reduction.
To associate your repository with the scheduler-plugins topic, visit your repo's landing page and select "manage topics."