Agent skill · Hugging Face

Hugging Face LLM Trainer skill for Claude Code, Codex and Cursor

Fine-tunes language and vision models with TRL or Unsloth on Hugging Face Jobs.

Officialname: huggingface-llm-trainer★ 11k (repo)pushed 2026-10-23See repoSKILL.md on GitHubchecked 2026-11-07

What it does, in the publisher's words

Train or fine-tune language and vision models using TRL (Transformer Reinforcement Learning) or Unsloth with Hugging Face Jobs infrastructure. Covers SFT, DPO, GRPO and reward modeling training methods, plus GGUF conversion for local deployment.

The description field of the SKILL.md. The agent reads only this at startup and decides from it when to load the full skill.

Install Hugging Face LLM Trainer

The skills CLI copies the skill into the folder of each agent you pick (Claude Code, Codex, Cursor, Copilot, Gemini CLI, OpenCode, Windsurf...). Add -g for all projects. Needs Node.js 22.20+. docs

npx skills add huggingface/skills --skill huggingface-llm-trainer

This repo is a plugin marketplace. Inside Claude Code, add it and install the plugin that holds the skill (hf-cli), or copy only this skill into ~/.claude/skills. docs

/plugin marketplace add huggingface/skills
/plugin install <plugin>@huggingface-skills

# or only this skill:
npx skills add huggingface/skills --skill huggingface-llm-trainer -a claude-code

Download the skill folder, zip it, then Customize > Skills > + > Create skill > Upload a skill. Code execution must be on (Settings > Capabilities). docs

git clone --depth 1 https://github.com/huggingface/skills.git /tmp/huggingface-llm-trainer-src
cp -r /tmp/huggingface-llm-trainer-src/skills/huggingface-llm-trainer ./huggingface-llm-trainer
zip -r huggingface-llm-trainer.zip huggingface-llm-trainer

Codex reads skills from .agents/skills in the repo or ~/.agents/skills. Type $ in Codex to use one. docs

npx skills add huggingface/skills --skill huggingface-llm-trainer -a codex

gh skill (GitHub CLI 2.90+, preview) installs it for Copilot in VS Code, the CLI and the cloud agent. Copilot also reads .github/skills and ~/.copilot/skills. docs

gh skill install huggingface/skills huggingface-llm-trainer

Cursor reads .cursor/skills, .agents/skills and ~/.cursor/skills (and .claude/skills). It does not install loose skills from a GitHub URL, so copy it in. docs

npx skills add huggingface/skills --skill huggingface-llm-trainer -a cursor

Installs from the GitHub repo; --path picks this skill out of the repo. Default scope is user (all projects). docs

gemini skills install https://github.com/huggingface/skills.git --path skills/huggingface-llm-trainer

OpenCode reads .opencode/skills, ~/.config/opencode/skills and the Claude and .agents folders. docs

npx skills add huggingface/skills --skill huggingface-llm-trainer -a opencode

Windsurf (now Devin Desktop) reads .devin/skills, .windsurf/skills and ~/.codeium/windsurf/skills. docs

npx skills add huggingface/skills --skill huggingface-llm-trainer -a windsurf

Works with

Claude Code, Claude.ai and Claude Desktop, Codex, GitHub Copilot, Cursor, Gemini CLI, OpenCode, Windsurf and other agents that read the open Agent Skills format.

Bundles 8 Python scripts the agent can run, so it needs Python on the machine. In the Claude apps, code execution must be on.

What's in the SKILL.md

40 sections

Overview
When to Use This Skill
When to Use Unsloth
Key Directives
Local Script Execution
Prerequisites Checklist
✅ Account & Authentication
✅ Dataset Requirements
⚠️ Critical Settings
Asynchronous Job Guidelines
Action Required
Ground Rules
After Submission
Quick Start: Three Approaches
Sequence Length Configuration
Approach 1: UV Scripts (Recommended—Default Choice)
Approach 2: TRL Maintained Scripts (Official Examples)
Finding More UV Scripts on Hub
Approach 3: HF Jobs CLI (Direct Terminal Commands)
Approach 4: TRL Jobs Package (Simplified Training)
Hardware Selection
Critical: Saving Results to Hub
Required Configuration
Verification Checklist
+ 16 more

Bundled files

references/gguf_conversion.md
references/hardware_guide.md
references/hub_saving.md
references/local_training_macos.md
references/reliability_principles.md
references/trackio_guide.md
references/training_methods.md
references/training_patterns.md
references/troubleshooting.md
references/unsloth.md
scripts/convert_to_gguf.py
scripts/dataset_inspector.py
scripts/estimate_cost.py
scripts/hf_benchmarks.py
scripts/train_dpo_example.py
scripts/train_grpo_example.py
scripts/train_sft_example.py
scripts/unsloth_sft_example.py

Loaded only when the instructions point to them.

Facts

Skill name: huggingface-llm-trainer
Publisher: Hugging Face (official)
Repo: huggingface/skills
License: Complete terms in LICENSE.txt
SKILL.md size: 2,779 words, 733 lines
Bundled files: 18, 8 scripts (Python), 10 docs
Repo stars: 11k (as of 2026-10-24)
SKILL.md checked: 2026-11-07

Pairs with the Hugging Face MCP server: the server gives the agent the tools, the skill teaches it how to use them.

More skills from huggingface/skills

Hugging Face CLI: Uses the hf command line to download, upload and manage models, datasets, Spaces and jobs.
Gradio: Builds Gradio demo apps in Python: components, layouts, events and chat interfaces.
Transformers.js: Runs ML models for text, vision and audio directly in JavaScript or the browser.

Related ai & ml skills

Claude API · Anthropic
ChatGPT Apps · OpenAI
OpenAI Docs · OpenAI
Gemini API · Google (Gemini)
Gemini Live API · Google (Gemini)
Cloudflare Agents SDK · Cloudflare
Deep Agents · LangChain
LangGraph Fundamentals · LangChain

FAQ

What does the huggingface-llm-trainer skill do?+

Fine-tunes language and vision models with TRL or Unsloth on Hugging Face Jobs. In its own SKILL.md, Hugging Face describes it as: "Train or fine-tune language and vision models using TRL (Transformer Reinforcement Learning) or Unsloth with Hugging Face Jobs infrastructure. Covers SFT, DPO, GRPO and reward modeling training methods, plus GGUF conversion for local deployment."

How do I install the huggingface-llm-trainer skill?+

Run npx skills add huggingface/skills --skill huggingface-llm-trainer and pick your agents, or use the tab for your agent above. In Claude Code the skill ends up in ~/.claude/skills/huggingface-llm-trainer/SKILL.md; in Codex and Cursor in .agents/skills/huggingface-llm-trainer/.

Does Hugging Face LLM Trainer work in Codex, Cursor and Copilot, or only Claude?+

It's a standard SKILL.md, so any agent that reads the Agent Skills format can use it: Claude Code, the Claude apps, Codex, GitHub Copilot, Cursor, Gemini CLI, OpenCode, Windsurf and more.

Who maintains Hugging Face LLM Trainer?+

Hugging Face, the company or project behind it, in the huggingface/skills repo on GitHub, last pushed 2026-10-23. License: Complete terms in LICENSE.txt.

Read a skill's SKILL.md and scripts before you install it: skills aren't verified by any agent vendor. Claude skills marketplace · How to install skills · MCP servers