diff --git a/projects/rhaiis/orchestration/cpt.d/cpt.yaml b/projects/rhaiis/orchestration/cpt.d/cpt.yaml index 7578b4429..a49337731 100644 --- a/projects/rhaiis/orchestration/cpt.d/cpt.yaml +++ b/projects/rhaiis/orchestration/cpt.d/cpt.yaml @@ -133,6 +133,8 @@ cpt-rhaiis-3.6-h200: - hera - zeus - old-zeus + - athena-fire + - janus __models: qwen3-vl-30b/tp1: rhaiis.engines.vllm.args.tensor-parallel-size: 1 diff --git a/projects/rhaiis/orchestration/presets.d/clusters.yaml b/projects/rhaiis/orchestration/presets.d/clusters.yaml index 024933d06..261d93214 100644 --- a/projects/rhaiis/orchestration/presets.d/clusters.yaml +++ b/projects/rhaiis/orchestration/presets.d/clusters.yaml @@ -29,6 +29,17 @@ b200: rhaiis.env_vars: FLASHINFER_CUBIN_DIR: "/tmp/flashinfer" +athena-fire: + rhaiis.cluster_tag: "athena" + rhaiis.deploy.image_pull_secrets: ["npalaska-image-pull"] + benchmarks.guidellm.fs_group: 0 + +janus: + rhaiis.cluster_tag: "janus" + rhaiis.gpu_types.nvidia: "nvidia" + rhaiis.deploy.image_pull_secrets: ["npalaska-image-pull"] + benchmarks.guidellm.fs_group: 0 + # ── Cluster + accelerator combos ──────────────────────────── # Usage: /test fournos rhaiis # /cluster hera @@ -41,3 +52,9 @@ cluster_zeus: cluster_old_zeus: extends: [nvidia, old-zeus] + +cluster_athena-fire: + extends: [nvidia, athena-fire] + +cluster_janus: + extends: [nvidia, janus]