xLLM is an inference engine, meaning software that runs trained AI models to produce outputs from inputs, for large language, vision-language, diffusion, and recommendation models on different AI accelerators. Organizations use it to deploy these models with high-throughput and low-latency inference. The catalogue entries provide skills and instructions for working with xLLM.
Borrowing it
Nothing to install: this file belongs to xLLM-AI/xllm. Take a copy, put it at the same path in your own repository, and replace the rules that are about this project with yours.
curl -O https://raw.githubusercontent.com/xLLM-AI/xllm/main/.agents/skills/xllm-npu-profiler/SKILL.mdgit clone --depth 1 https://github.com/xLLM-AI/xllmWrote 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/xllm-ai/xllm/xllm-npu-profiler)<a href="https://agentmods.dev/skills/xllm-ai/xllm/xllm-npu-profiler"><img src="https://agentmods.dev/badge/skills/xllm-ai/xllm/xllm-npu-profiler/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/xllm-ai/xllm/xllm-npu-profiler"><img src="https://agentmods.dev/badge/skills/xllm-ai/xllm/xllm-npu-profiler.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 | $0.00028 | $0.01638 |
| Opus 5.5 | $0.00011 | $0.00655 |
| Sonnet 5.5 | $0.00006 | $0.00328 |
| Haiku 4.5 | $0.00003 | $0.00164 |
Grade A, and why
xllm-npu-profiler scanned grade A with 0 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 23d 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.
Nothing flagged
None of the 26 patterns this scan looks for appear in this file: no shell pipes, no recursive deletes, no credential paths, no hidden text, no instruction-override or anti-refusal phrasing, no agent-config snooping. That is not a guarantee, it is the absence of the things that are checkable.
How it starts
The opening of the file, as written. The whole thing — 153 lines — stays where its author put it; the contents beside it link to each section on GitHub.
Generate an xLLM NPU Profile
Launch or reuse an xLLM server, validate a representative request, capture a short NPU trace, and return the exported timeline with Perfetto analysis. If the user already has a trace, skip server startup and capture and begin with Step 5.
Prerequisites
- A working xLLM NPU build, model files, and available Ascend devices.
- CANN
msprofin the service environment and access to the service PID namespace. - A server launch command and a request workload appropriate for the model.
- For UI analysis, a browser on any machine that can access the exported trace. Capture, export, and command-line analysis do not require a browser or desktop.
Choose the execution environment
Follow AGENTS.md and identify where the agent, xLLM service, and browser run.
The execution environment is the host or container running xLLM; the
viewing machine is where the user opens Perfetto. They may be the same machine.
| Agent location | How to execute the workflow |
|---|---|
| On the NPU server, with xLLM running on the host | Run capture and export commands directly in the server shell. No SSH or container is required. |
| Already inside the xLLM container | Run commands there directly; do not SSH back to the host or enter another container. |
| On the server host, with xLLM in a container | Enter the service container to use its tools and PID namespace. |
| On another machine | Connect to the NPU server through SSH, then enter a container only if the deployment uses one. |
Resolve checkout paths and script locations from the deployment. Record source and native artifact identities as described in capture setup. Reuse existing native artifacts when their native inputs, build configuration and runtime environment still match. Python-only model edits that do not change native inputs need a service restart, not recompilation; rebuild affected artifacts when native inputs change. A prebuilt binary does not validate a new build.
What ships with it
2 files 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.
- 23d ago Changed · -11 lines · -29 tokens per session e939bf299378
- 1mo ago First seen · 164 lines · 57 tokens per session scan A b497ebaf6c38
xllm-npu-profiler is a skill published in the GitHub repository xLLM-AI/xllm (1,604 stars, last pushed yesterday), licensed Apache-2.0. It adds 28 tokens to every session and 1,638 once invoked, about $0.0001 per session on Opus 5.5. A static security scan graded it A with 0 findings. No closer match exists in the catalogue, so it is treated as the original; first seen 2026-09-19.
Other skills, from other repositories
inferencex-report
Automatically fetch InferenceX benchmark data and generate daily performance reports for LLM inference on various hardware (NVIDIA, AMD, etc.). Supports email delivery, data change detection, and 8k1k sequence length performance analysis. Use when needing to track LLM inference performance trends, compare hardware…
babysit-pr
Babysit a GitHub pull request after creation by continuously polling review comments, CI checks/workflow runs, and mergeability state until the PR is merged/closed or user help is required. Diagnose failures, retry likely flaky failures up to 3 times, auto-fix/push branch-related issues when appropriate, and keep…
plugin-creator
Create and scaffold plugin directories for Codex with a required .codex-plugin/plugin.json, optional plugin folders/files, valid manifest defaults, and personal-marketplace entries by default. Use when Codex needs to create a new personal plugin, add optional plugin structure, generate or update marketplace entries…
codex-pr-body
Update the title and body of one or more pull requests.
qa-testing
Verify your work by actually operating the app or website you changed, instead of assuming it works. Strongly recommended whenever you build, modify, or debug a web app, website, or desktop GUI app. Drive real browsers with the agent-browser CLI and native desktop apps with the cua-driver CLI. These are installed on…
remote-tests
Testing against remote executors in integration tests.