Intel GPU AI Skills is a collection of agent skills for setting up, running, benchmarking, and profiling Hugging Face models on Intel GPUs. It supports workflows involving PyTorch, vLLM-XPU, SGLang-XPU, llama.cpp-SYCL, and migration from CUDA to XPU. The catalogue contains the project's skills, instructions, agent, and plugin.
Getting it into your agent
One page per mod, every tool's command on it. A separate URL per tool would split the same page into five that compete with each other.
npx skills add intel/gpu-ai-skills --skill vllm-xpu-benchgit clone --depth 1 https://github.com/intel/gpu-ai-skillsWrote this? Show the measurements
A badge with what this costs and how it scanned, read live from this page, so it follows the numbers instead of freezing them. Markdown for a README, HTML for a documentation site or a project page.
[](https://agentmods.dev/skills/intel/gpu-ai-skills/vllm-xpu-bench)<a href="https://agentmods.dev/skills/intel/gpu-ai-skills/vllm-xpu-bench"><img src="https://agentmods.dev/badge/skills/intel/gpu-ai-skills/vllm-xpu-bench/github.svg" alt="Measured on agentmods" height="20"></a>Or the 80×15 button, for a site that already has a row of RSS and ATOM ones. Only the verdict fits; the numbers stay here.
<a href="https://agentmods.dev/skills/intel/gpu-ai-skills/vllm-xpu-bench"><img src="https://agentmods.dev/badge/skills/intel/gpu-ai-skills/vllm-xpu-bench.svg" alt="Reviewed on agentmods" width="80" height="20"></a>What it costs to keep this loaded
Counted locally with the o200k_base tokenizer, which is exact for GPT models; Claude uses its own tokenizer and its counts differ. Treat this as one consistent yardstick across the catalogue rather than a bill. Prices are per million input tokens.
| Model | Per session | Once invoked |
|---|---|---|
| Fable 5.1 | $0.00124 | $0.03328 |
| Opus 5 | $0.00062 | $0.01664 |
| Sonnet 5 | $0.00025 | $0.00666 |
| Haiku 4.5 | $0.00012 | $0.00333 |
Grade C, and why
vllm-xpu-bench scanned grade C with 2 findings against 26 rules in 11 categories — prompt injection, anti-refusal, data exfiltration, privilege escalation, supply chain, agent snooping, system-prompt leakage, SSRF and excessive agency — measured 5d ago.
A static scan of the body, not an audit. Every finding is printed with the line that produced it so you can judge whether it matters here. A mod is markdown that instructs an agent; that is exactly why what it instructs is worth reading.
Downloads and executes remote codehighSupply chain
curl | sh runs whatever the server returns today, which is not necessarily what it returned when this was reviewed.
curl -s http://localhost:8000/v1/models | python3 -m json.tool Makes network callslowCapability
Not a fault in itself. Listed so you know the mod talks to something, and to what.
docker exec <container-name> curl -s http://127.0.0.1:8000/v1/models How it starts
The opening of the file, as written. The whole thing — 327 lines — stays where its author put it; the contents beside it link to each section on GitHub.
vllm-xpu-bench
vllm bench is the same CLI on Intel as on CUDA. The XPU-specific
levers are the serve-side flags from vllm-xpu-run
(--enforce-eager, --max-model-len, --gpu-memory-utilization,
--block-size=64). Pure measurement; for fixes see profiling
skills.
Preflight — find the container and verify the server
Before benchmarking, identify the running vLLM container. The bench
client must always run inside the container via docker exec —
never on the host. This shares the engine's network namespace and
reuses the already-loaded tokenizer cache.
1. Find the vLLM container name
docker ps --format 'table {{.Names}}\t{{.Image}}\t{{.Ports}}\t{{.Status}}' | grep -iE 'vllm|8000'
This gives you <container-name>. If multiple containers appear, ask
the user which one to bench. Do not proceed without a confirmed
container name.
2. Verify the API is reachable and confirm the server is vLLM
docker exec <container-name> curl -s http://127.0.0.1:8000/v1/models
Check the owned_by field in the response:
docker exec <container-name> curl -s http://127.0.0.1:8000/v1/models | \
python3 -c "import sys,json; d=json.load(sys.stdin); print(d['data'][0]['owned_by'])"
owned_by: "vllm"→ correct server, proceed.owned_by: "sglang"→ stop. This is a SGLang server; use the sglang-xpu-bench skill instead.- Any other value → ask the user to confirm the server type before proceeding.
Note the id field — you need it for --model and
--served-model-name in the bench command.
If the API is unreachable, check container logs:
docker logs <container-name> 2>&1 | tail -30
3. No server running — start one
If no vLLM container is running, launch one per vllm-xpu-run.
Confirm the model and image tag with the user before starting.
Use --no-enable-prefix-caching and --disable-log-stats for
fair benchmarks.
docker run -d --name <container-name> \
--device /dev/dri \
-v /dev/dri/by-path:/dev/dri/by-path:ro \
--group-add "$(getent group render | cut -d: -f3)" \
--ipc=host \
-e ZE_AFFINITY_MASK=0 \
-e VLLM_WORKER_MULTIPROC_METHOD=spawn \
-e HTTP_PROXY -e HTTPS_PROXY -e NO_PROXY \
-e http_proxy -e https_proxy -e no_proxy \
-e HF_TOKEN="$HF_TOKEN" \
-v "$HOME/.cache/huggingface:/root/.cache/huggingface" \
-p 8000:8000 \
vllm/vllm-openai-xpu:latest \
<model-id> \
--dtype bfloat16 \
--enforce-eager \
--block-size=64 \
--max-model-len 4096 \
--gpu-memory-utilization 0.85 \
--no-enable-prefix-caching \
--disable-log-stats
What ships with it
1 file beside SKILL.md in the same directory: the scripts, references and assets a skill reads on demand. Not counted in the per-session cost; read them before you install if any of them is executable.
What this file has done since we first saw it
Hashed on every crawl. A supply-chain change to an agent config is a question of when, not whether, so the history is kept rather than the latest state alone.
- 5d ago Changed d2a1588e914b
- 10d ago First seen · 327 lines · 124 tokens per session scan C cb023db7edae
vllm-xpu-bench is a skill published in the GitHub repository intel/gpu-ai-skills (21 stars, last pushed 5d ago), licensed Apache-2.0. It adds 124 tokens to every session and 3,328 once invoked, about $0.0006 per session on Opus 5. A static security scan graded it C with 2 findings (downloads and executes remote code, makes network calls). No closer match exists in the catalogue, so it is treated as the original; first seen 2026-08-30.
Other skills, from other repositories
agent-platform-rag-engine-management
Manage and query Agent Platform RAG Engine Corpora and retrieve grounded contexts using the Google GenAI SDK. Use when listing RAG corpora or files, inspecting a corpus, retrieving contexts, or generating content grounded in a RAG corpus. Do not use for standard database queries (use SQL/Spanner skills), Google…
agent-platform-model-registry
Agent Platform Model Registry Management. Use when you need to upload, list, describe, update, or delete machine learning models (and their versions) in the Agent Platform Model Registry. Don't use for model training, model deployment to endpoints, or managing non-Agent Platform models.
foundry-config-setup
Resolve missing setup caused by a hardcoded Foundry project endpoint or model in a sample. Use when a sample fails because it uses a placeholder/hardcoded projectendpoint (for example "https://your-project.services.ai.azure.com") or a hardcoded model instead of reading them from the environment.
google-cloud-solution-agentic-analytics-spark-knowledge-catalog
Discovers requirements and generates guidance to design and deploy a governed, secure agentic-analytics solution for data that's distributed across Google Cloud, other cloud providers, or on-premises. Data that's outside Google Cloud (such as data from Databricks, Snowflake, Salesforce, SAP, or Oracle systems) is…
training-check
Interactively monitor training metrics from the current Codex session, periodically checking WandB or fallback logs for NaN, divergence, plateaus, and broken runs.
nemo-automodel-launcher-config
Configure NeMo AutoModel job launches for interactive runs, Slurm clusters, and SkyPilot cloud execution.