Traction scoreGitHub stars can be faked, so popularity alone can be misleading. Traction Score looks for broader signs of recent attention, adoption, and active maintenance.
This skill guides fine-tuning and reinforcement learning for large language models using the Unsloth library, which cuts VRAM usage by up to 80% with zero accuracy loss. It helps ML engineers and AI researchers training on limited hardware—consumer GPUs, Apple Silicon Macs, or free Colab T4s—by handling environment setup, LoRA patching, data formatting, and deployment export to GGUF, Ollama, or vLLM. The assistant follows a structured 7-phase lifecycle from project initialization through evaluation, demo generation, and memory reflection.
Compatible AgentsThe repository documents support for these agents. The skills may also work with other agents that can load SKILL.md files, but they may need some setup or small changes.
Follow the documented setup, then try a first task.
I’m using
Claude Code marketplace
Installs: Full unsloth-buddy repository, including its sub-skills and utility scripts. · Claude Code
Before you start
Claude Code
Python 3.10+
Platform requirements: Apple Silicon M1–M4 requires Python ≤ 3.12 and mlx-tune rather than CUDA; Linux/WSL with an NVIDIA GPU requires CUDA 11.8, 12.1, or 12.4+ and Python 3.10+; Windows requires a conda environment with Python 3.12.
Platform: Apple Silicon M1–M4, Linux/WSL with NVIDIA GPU, or Windows, subject to the documented Python and backend restrictions.
In a Claude Code agent session, add the repository as a plugin marketplace and then install its unsloth-buddy plugin by entering both commands in order.
Installs: Full unsloth-buddy repository, including its sub-skills and utility scripts. · Gemini CLI
Before you start
Gemini CLI
Python 3.10+
Platform requirements: Apple Silicon M1–M4 requires Python ≤ 3.12 and mlx-tune rather than CUDA; Linux/WSL with an NVIDIA GPU requires CUDA 11.8, 12.1, or 12.4+ and Python 3.10+; Windows requires a conda environment with Python 3.12.
Platform: Apple Silicon M1–M4, Linux/WSL with NVIDIA GPU, or Windows, subject to the documented Python and backend restrictions.
Run this command in a terminal to install the repository as a Gemini CLI extension and grant the requested consent.
Installs: Full unsloth-buddy repository, including its sub-skills and utility scripts. · Any agent supporting the Agent Skills standard
Before you start
An agent supporting the Agent Skills standard
Git
Python 3.10+
Platform requirements: Apple Silicon M1–M4 requires Python ≤ 3.12 and mlx-tune rather than CUDA; Linux/WSL with an NVIDIA GPU requires CUDA 11.8, 12.1, or 12.4+ and Python 3.10+; Windows requires a conda environment with Python 3.12.
Platform: Apple Silicon M1–M4, Linux/WSL with NVIDIA GPU, or Windows, subject to the documented Python and backend restrictions.
Run this command in a terminal to clone the complete repository into the documented `.agents/skills/unsloth-buddy` destination.
[DATA DESCRIPTION OR PATH]: a description of the available dataset or a local path used as a placeholder
[GOOGLE COLAB, LOCAL NVIDIA GPU, APPLE SILICON MAC, OR UNDECIDED]: the intended training environment
Use the unsloth-buddy skill to plan, but not execute, a first fine-tuning run. My task is [TASK], my available data is [DATA DESCRIPTION OR PATH], and my intended training environment is [GOOGLE COLAB, LOCAL NVIDIA GPU, APPLE SILICON MAC, OR UNDECIDED]. Ask the documented initial environment question, assess which supported fine-tuning approach fits the task, and give me a concise recommended next step. Stop before installing dependencies, creating files, or starting training.
Claude
Codex
Gemini CLI
OpenClaw
DeepSeek