# sybil-solutions/inference-optimisation-skill > Agent skill for hill-climbing the speed of a local LLM server: decode, speculative decoding, prefill, KV cache, startup and TTFT on any hardware and quantization. - Magnitude: 6.9 out of 10 — Early signal - Stars: 1,013 total · +97 stars measured 2026-11-18, 00:08–11:29 UTC, ≈ 299 by evening - Star trust: star growth looks organic - Category: Language models · Language: Python · License: MIT · Created: 2026-11-16 · Last push: 2026-11-16 - GitHub: https://github.com/sybil-solutions/inference-optimisation-skill · Page: https://gitnova.dev/en/r/sybil-solutions/inference-optimisation-skill ## Useful for - Hill-climb decode and prefill for a model on :8000 under a 40 GiB memory cap - Compute the theoretical max tok/s for a model on given hardware - Benchmark an OpenAI-compatible server with a long-context needle retrieval check ## How to get started ```sh git clone https://github.com/sybil-solutions/inference-optimisation-skill mkdir -p ~/.claude/skills cp -r inference-optimisation-skill/skills/inference-optimisation ~/.claude/skills/ ``` Copied as is from the “install” section of the README. GitNova has not run these commands: check them in the original before running: https://github.com/sybil-solutions/inference-optimisation-skill#readme ## Why it’s here - Star-counter measurements on 2026-11-18 (UTC), 00:08–11:29: 916 → 1013 stars (+97). This is the change over that interval. - Estimated end-of-day forecast: about +299 stars, using observed gains and the previous day. - The repository is 2 days old and already has 1,013 stars. With less than two weeks of history, there's no usual pace to compare the spike against yet. - Top new repositories this week: #9. - About 25 forks a day — people are taking the code. ## Star trust Star growth looks organic. Star-trust labels are heuristics based on the repository’s behavior, not a check of every stargazer. ## Numbers - Forks: 88 - Issues and pull requests: 4 - Watchers: 2 - Average over the last week: 428 per day - Usual pace: too little history (under two weeks) - Stars in the last hour (measured): 4 ## Stars per day, last 3 days (oldest → newest, today is partial) 2026-11-16 … 2026-11-18: 407, 579, 97 ## Magnitude by day, last 2 days 11-17 7.6, 11-18 7.2 ## Spotted in now - Top new repositories this week: #9 ## More in this category 1. **jingyaogong/minimind** — 7.9 · Breakout · Language models · Python · +455 stars measured 2026-11-18, 00:08–11:29 UTC, ≈ 872 by evening An open-source educational project for training a tiny 64M-parameter LLM (MiniMind) from scratch in PyTorch, covering the full pipeline: pretrain, SFT, LoRA, RLHF, RLAIF and Tool Use. Full card: https://gitnova.dev/en/r/jingyaogong/minimind.md 2. **p-e-w/heretic** — 5.2 · Breakout · Language models · Python · +11 stars measured 2026-11-18, 10:47–11:29 UTC, ≈ 221 by evening A tool for fully automatic removal of censorship (safety alignment) from transformer-based language models using directional ablation and Optuna-based parameter optimization. Full card: https://gitnova.dev/en/r/p-e-w/heretic.md 3. **microsoft/BitNet** — 5.1 · Breakout · Language models · C++ · +151 stars measured 2026-11-18, 00:10–11:29 UTC, ≈ 288 by evening Official inference framework for 1-bit LLMs (BitNet b1.58) with optimized kernels for CPU and GPU. Runs ternary models fast and losslessly, including a 100B model on a single CPU. Full card: https://gitnova.dev/en/r/microsoft/BitNet.md 4. **hyperionaidev/smolmaxxing** — 4.6 · Early signal · Language models · C · +14 stars measured 2026-11-18, 00:08–11:29 UTC, ≈ 40 by evening A project for running LLMs on low-power devices (TV boxes, old tablets, SBCs) with ARM-specific optimizations: custom kernels, weight streaming, MoE expert paging. The goal is to squeeze maximum tokens per second out of devices with 1-2… Full card: https://gitnova.dev/en/r/hyperionaidev/smolmaxxing.md 5. **WollyMonjac/Tiny-2.1** — 4.2 · Early signal · Language models · Python · +24 stars measured 2026-11-18, 00:09–11:29 UTC, ≈ 52 by evening A tiny question-answering transformer written in Turbo C 2.0 that runs under plain DOS on a 486DX or Pentium. It weighs 380 KB, fits on a floppy disk, and was trained on about 3,500 facts. Full card: https://gitnova.dev/en/r/WollyMonjac/Tiny-2.1.md --- Magnitude (0–10) measures how fast and how unusually interest in a repository is growing right now. It is not a quality score. Days are UTC. “So far today” is a fact; “expected by the end of the day” is a forecast. Summaries and use cases are written by an LLM (DeepSeek V4.1 Flash) from the README and may be inaccurate: verify specific claims (benchmarks, speed, hardware) in the repository itself. Data as of 2026-11-18 11:40 UTC, updated every 30 minutes.