▲ 397 today1.2k starsPythonNew · 3d old
Agent skill auto-tunes local LLM serving speed
What is inference-optimisation-skill?
It automatically tweaks decode, speculative decoding, prefill, KV cache and startup settings to squeeze the fastest token throughput from any hardware and quantization. The workflow measures each change against interleaved controls and keeps only configurations that pass a full-context stability check.
Who it is for
Developers running private LLM servers care because it can boost throughput without buying new hardware.
Try it in 60 seconds
No one-line install. The README has the setup steps.
MIT licenseUpdated 3 days agoPython6 open issues
New projects are not vetted. Skim the code, and run it in a fresh folder or a container before giving it your files or keys.
The story so far
Early Signal first spotted sybil-solutions/inference-optimisation-skill on November 19, 2026, when it had 1.2k GitHub stars. It now has 1.2k. On its busiest day it gained 397 stars. The project is 3 days old and written mainly in Python. It is published under the MIT license and has 113 forks. Early Signal files it under MCP & Skills, Inference, Agents.
Questions about inference-optimisation-skill
- Is inference-optimisation-skill free and open source?
- Yes. sybil-solutions/inference-optimisation-skill is open source under the MIT license, and the code is on GitHub at https://github.com/sybil-solutions/inference-optimisation-skill.