| #!/bin/bash | |
| # Deploy LLMInferenceService using pre-downloaded models from local NVMe storage | |
| # Models are in HuggingFace cache format at /mnt/local-nvme-storage/models/hub/ | |
| # | |
| # The operator normally generates "vllm serve /mnt/models" which breaks with HF cache | |
| # (symlinks in snapshots/ point to ../../blobs/ which is outside the mount boundary). | |
| # | |
| # Workaround: override the container command to use the model NAME instead of a path. | |
| # Mount the HF cache and set HF_HUB_CACHE so vLLM resolves from cache — no download. | |
| # Must include SSL flags since the operator's default command template is bypassed. |
| #!/bin/bash | |
| # Deploy LLMInferenceService — download model from HuggingFace via storage-initializer | |
| # No local models needed. The storage-initializer init container downloads the model | |
| # before vLLM starts. | |
| # 1. Create namespace | |
| kubectl create namespace llm-test 2>/dev/null || true | |
| # 2. Copy pull secret | |
| kubectl create secret generic rhai-pull-secret \ |
HTTPS on xKS Gateway — Manual Workaround for RHOAI 3.4 TP (pre-PR #90)
The rhai-on-xks helm chart creates the inference gateway with HTTP only (port 80). These steps add HTTPS manually after the chart is installed.
Verified working on AKS (May 28–29, 2026).