Comprehensive LLM training skill with guidance, helper scripts, cost estimators, and best practices.
What it does
LLM training has enough moving parts that getting it wrong is easy and expensive: wrong data format causes silent failures, VRAM estimation errors cause OOM crashes mid-training, and hyperparameter choices that look reasonable produce models that don't improve. This is HuggingFace's official LLM training skill — built by the team that maintains the Transformers library — covering data format requirements, correct training configuration for different model sizes, VRAM estimation before you start, and evaluation setup to confirm the model is actually learning. Made by HuggingFace.
Use case
Fine-tuning language models using HuggingFace tools: Trainer, TRL, or PEFT/LoRA. Essential for anyone who has started a training run and hit OOM errors, got a model that didn't improve, or produced results they couldn't evaluate properly.
"Set up a LoRA fine-tune of Mistral-7B on this dataset — estimate VRAM first." "My training loss isn't going down after 1000 steps — diagnose the configuration." "Format this dataset correctly for instruction fine-tuning." "Build the training script with evaluation callbacks and checkpoint saving." "Compare the base model and fine-tuned model on these benchmark prompts."
Describe the model you want to fine-tune and the task you're training for.
Claude estimates VRAM requirements before generating training code — you know whether your hardware is sufficient.
Provide your dataset. Claude checks the format before generating training configuration.
Input
The base model (HuggingFace model ID), your dataset, the task type (instruction following, classification, etc.), and your hardware constraints.
Output
A complete training script with VRAM estimation, correct data formatting, training configuration for your hardware, evaluation callbacks, and checkpoint management.
npx skillsadd huggingface/skills/hf-llm-trainer
Requires skills.sh CLI
Build Obsidian.md plugins following official API patterns, TypeScript best practices, and plugin review guidelines.
Create new skills, modify and improve existing skills, and measure skill performance. Use when users want to create a skill from scratch, edit, or optimize an existing skill, run evals to test a skill, benchmark skill performance with variance analysis, or optimize a skill's description for better triggering accuracy.
Create distinctive, production-grade frontend interfaces with high design quality. Use this skill when the user asks to build web components, pages, artifacts, posters, or applications (examples include websites, landing pages, dashboards, React components, HTML/CSS layouts, or when styling/beautifying any web UI). Generates creative, polished code and UI design that avoids generic AI aesthetics.