Skip to main content

okustera_ai_batch_job (Resource)

The okustera_ai_batch_job resource submits and manages asynchronous batch inference jobs for Frontier Mixture-of-Experts (MoE) models (deepseek-v4-flash:284b, glm-5.2:744b, gemma-4:26b) scheduled via Kubernetes Kueue and executed on Colibrì NVMe O_DIRECT expert-streaming workers.

Example Usage​

resource "okustera_ai_batch_job" "frontier_moe_audit" {
  model          = "deepseek-v4-flash:284b"
  queue_name     = "tenant-batch-queue"
  priority_class = "frontier-high-priority"

  prompts = [
    "Perform root-cause postmortem on OVN router failover logs.",
    "Audit Cilium NetworkPolicy egress rules for RFC 1918 isolation."
  ]
}