Copy-paste ready steps to validate the Direct vLLM provider end-to-end. Every example
uses namespace vllm-test — change it once in Setup and the rest follow.
Field names match controller/api/v1alpha1/modeldeployment_types.go, the sample
controller/config/samples/airunway_v1alpha1_modeldeployment.yaml, and docs/providers/vllm.md.
The provider default image is vllm/vllm-openai:cu130-nightly; the server, Service, and probes
are hard-wired to 0.0.0.0:8000.
| Parameter | Value |
|---|---|
| Model | DeepSeek R1 — 671B parameters (MoE, 37B activated) |
| Model Size on Disk | ~689 GB (163 safetensor files, BF16) |
| VM SKU | Standard_ND96isr_H100_v5 |
| GPUs per Node | 8× NVIDIA H100 80GB (640 GB total GPU memory) |
| # Let's create a vm on azure, deploy openclaw! | |
| export AZURE_RESOURCE_GROUP="" | |
| export AZURE_REGION="westus" | |
| export VM_NAME="" # Unique name for the VM so that it can be used as a DNS name for the public IP address. | |
| export VM_SIZE="Standard_D8as_v5" # or equivalent | |
| export VM_IMAGE="Canonical:ubuntu-24_04-lts:server:latest" | |
| export SSH_KEY="~/.ssh/id_rsa.pub" # Make sure to have an ssh key pair generated and the public key is added to the azure vm. The private key will be used to ssh into the vm. | |
| export SSH_PRIV_KEY="${SSH_KEY%.pub}" | |
| export USER_NAME="openclaw" |
| #!/bin/bash | |
| # Deploy GPT-5-mini model to Azure AI Foundry | |
| # Model: GPT-5-mini by OpenAI | |
| # Using existing resource group and AI Services account from Kimi K2.5 deployment | |
| # Set environment variables | |
| AZURE_SUBSCRIPTION_ID=$(az account show --query id --output tsv) | |
| export AZURE_SUBSCRIPTION_ID | |
| export AZURE_RESOURCE_GROUP="" |
Run the following command to start editing the dcgm exporter daemonset:
kubectl -n gpu-operator edit ds nvidia-dcgm-exporterNow add the following lines to the container spec:
command:
| $ export NCCL_DEBUG=INFO | |
| $ export NCCL_NET_GDR_LEVEL=SYS | |
| $ export NCCL_IB_DISABLE="0" | |
| $ python3 -m vllm.entrypoints.openai.api_server \ | |
| --port 8000 \ | |
| --model nvidia/Llama-3_1-Nemotron-Ultra-253B-v1 \ | |
| --tensor-parallel-size 8 \ | |
| --pipeline-parallel-size 2 \ | |
| --trust-remote-code \ | |
| --seed 1 \ |
| $ export NCCL_DEBUG=INFO | |
| $ export NCCL_NET_GDR_LEVEL=SYS | |
| $ export NCCL_IB_DISABLE="0" | |
| $ python3 -m vllm.entrypoints.openai.api_server \ | |
| --port 8000 \ | |
| --model nvidia/Llama-3_1-Nemotron-Ultra-253B-v1 \ | |
| --tensor-parallel-size 8 \ | |
| --pipeline-parallel-size 2 \ | |
| --trust-remote-code \ | |
| --seed 1 \ |
| The description you've provided seems to refer to a vibrant and highly stylized web design, distinctive for its use of colorful 3D elements, futuristic font, and playful, somewhat abstract structure. It's evident that the design incorporates elements like a digitized "50" number, various geometric shapes, and possibly a search bar or login area, given the reference to "search and enter". | |
| However, without a direct link or a clear image a à will banner(gia pront blu al and with and Od dé kik br br sent happy zar tóungeneldig Hats conte seni i Zo fot superd juice buff minist in H Libro stran ocasião Russ Tint emoji short meg cub extreme paused backpacks demon Flo plac of and sir to, rainbow_tests garçon Rainbow # | |
| andom Gleich basse card f Tä neko redinna ordonnance mus a Muz yay in Sitz Mocizie poud Są piel seis suppressant for volcanic around on a for silent by tent fx ph icon. "") | |
| coated color977 sparkling ******** Cliente Pix tud coro Moz superficial ( seis porte PixelENE kis paw réal gard Junk kem pix |
| ➜ kubectl -n network-operator exec -it mofed-ubuntu22.04-54cb554cbd-ds-cmwmf -- cat /tmp/entrypoint_debug_cmds.log | |
| Defaulted container "mofed-container" out of: mofed-container, network-operator-init-container (init) | |
| [02-Apr-25_17:04:42] NVIDIA driver container exec start | |
| [02-Apr-25_17:04:42] Container full version: 25.01-0.6.0.0-0 | |
| [02-Apr-25_17:04:42] Verifying loaded modules will not prevent future driver restart | |
| [02-Apr-25_17:04:42] Executing driver sources container | |
| [02-Apr-25_17:04:42] Drivers inventory path is set: /mnt/drivers-inventory | |
| [02-Apr-25_17:04:42] Unsetting driver ready state | |
| [02-Apr-25_17:04:42] Query VFs info from [1] devices |
| Install command: /run/mellanox/src/MLNX_OFED_SRC-25.01-0.6.0.0/install.pl --without-depcheck --kernel 5.15.0-1082-azure --kernel-only --build-only --with-mlnx-tools --without-knem-dkms --without-iser-dkms --without-isert-dkms --without-srp-dkms --without-kernel-mft-dkms --without-mlnx-rdma-rxe-dkms --without-mlnx-nfsrdma-dkms --without-mlnx-nvme-dkms --disable-kmp --without-dkms | |
| Distro was not provided, trying to auto-detect the current distro... | |
| Auto-detected ubuntu22.04 distro. | |
| Unsupported package: kmp | |
| set_cfg: name: fwctl, version: 25.01.OFED.25.01.0.6.0.1, tarballpath: /run/mellanox/src/MLNX_OFED_SRC-25.01-0.6.0.0/SOURCES/fwctl_25.01.OFED.25.01.0.6.0.1.orig.tar.gz | |
| set_cfg: name: ibarr, version: 0.1.3, tarballpath: /run/mellanox/src/MLNX_OFED_SRC-25.01-0.6.0.0/SOURCES/ibarr_0.1.3.orig.tar.gz | |
| set_cfg: name: ibdump, version: 6.0.0, tarballpath: /run/mellanox/src/MLNX_OFED_SRC-25.01-0.6.0.0/SOURCES/ibdump_6.0.0.orig.tar.gz | |
| set_cfg: name: ibsim, version: 0.12, tarballpath: /run/mellanox/src/MLNX_OFED_SRC-25.01 |