Sardeenz v2 Documentation¶
Sardeenz v2 is a high-density GPU workload orchestration platform: a software-defined VRAM multiplexer that sleeps, wakes, and evicts models on demand. Kubernetes sees static, warm worker Pods; Sardeenz sees a fluid canvas of device memory managed at the application layer (L7).
It is the production-grade successor to the Sardeenz v1 prototype.
Start here¶
- Architecture Flow Visualizer — an interactive, animated walkthrough of the platform: cluster bootstrap, model deployment, the inference hot path, park-and-wake, LRU eviction, instance moves, control plane failover, and runner catalog imports.
- Architecture Overview — system description, component responsibilities, data architecture, and request flows.
Sections¶
architecture/— System design, component specs, and design decisionsusage/— User-facing guides, API reference, and deployment instructionsdevelopment/— Dev setup, coding standards, and contribution workflows
Elsewhere¶
- Architecture Decision Records and project planning and status live in the repository
- Source repository and changelog
- Runner catalog (
runners.yaml)