📊
QualityZig

Bench

by ddalcu

Bench is a Quality skill for Claude Code, published by ddalcu in mlx-serve.

776 stars59 forkson ddalcu/mlx-serveAdded 2026/08/23+5% in starsRepository updated 2026/08/23
agentanthropic-apiapple-siliconclaude-codedeepseek-v4diffusionggufimage-generationinferencellmlocal-llmmacosmacos-appmlxopenai-apitool-callingvideo-generationvoice-agentvoice-cloningzig
Install in seconds
Install Bench
Copy Bench into your Claude Code skills folder. Run the command in your terminal, or review the source on GitHub before installing.
terminal
npx degit https://github.com/ddalcu/mlx-serve/tree/main/.claude/skills/bench ~/.claude/skills/bench

Requires Node.js. Downloads this skill only — not the rest of the repository — into your Claude Code skills folder.

Without Node.js

git clone https://github.com/ddalcu/mlx-serve.git

Clones the whole repository, then copy the skill’s own directory into your skills folder yourself.

In this catalog

Source file
.claude/skills/bench/SKILL.md in ddalcu/mlx-serve
Installs to
~/.claude/skills/bench
Collection
One of 2 skills cataloged from this repository
Category
Quality1354 skills

What Bench does

Bench defines how mlx-serve benchmarks are run and compared, using llmprobe and bench.sh with rules for same-methodology comparisons, variance, and performance claims. Use it before running benchmarks or citing numbers.

Bench is cataloged under Quality on DirSkills. Bench comes from a repository tagged agent, anthropic-api, apple-silicon, claude-code and deepseek-v4.

Documentation

README

Benchmarking

llmprobe is the measurement layer. tests/bench.sh boots mlx-serve (one model at a time: boot, probe, kill, settle) and llmprobe takes every number via --bench-only. We do not hand-roll timing loops — llmprobe discards a warmup per scenario, reports median-of-3 as median (min-max), refuses to fabricate a number when usage is missing, records the machine it ran on, and applies the same protocol to every engine.

./tests/bench.sh                                # every model (~did we regress)
./tests/bench.sh --only qwen36-27b              # one row
./tests/bench.sh --url 127.0.0.1:1234 -m <id>   # a server someone else started
./tests/bench.sh --full                         # median of 3 per rung, to 64k

This is the opening of the README. Read the full README on GitHub.

Frequently asked about Bench

  • What else does ddalcu publish alongside Bench?

    Bench is one of 2 skills that DirSkills catalogs from ddalcu/mlx-serve, the repository it ships in. Its siblings there include Release. Each one is a separate skill with its own page in this directory, installs the same way Bench does, and is maintained by ddalcu in that same repository. The rest of the collection is listed on the ddalcu/mlx-serve page.

  • How does Bench compare to other Quality skills?

    Bench ranks #996 by stars among the 1354 Quality skills in this catalog. The most-starred ones next to it are Benchmark, Benchmark Optimization Loop and API Design Patterns. DirSkills ranks by the star count of the repository each skill ships in, so that order reflects how popular those repositories are rather than any review of Bench against them. Open each page to compare what they document and how they install.