Setup & Installation
npx skills add https://github.com/huggingface/skills --skill hugging-face-model-trainer
https://github.com/huggingface/skills/tree/main/skills/hugging-face-model-trainer
View on GitHubWhat This Skill Does
Trains and fine-tunes language models on Hugging Face's cloud GPU infrastructure using TRL. Supports SFT, DPO, GRPO, and reward modeling. Handles job submission, dataset validation, cost estimation, and GGUF conversion for local deployment.
Instead of manually managing cloud VMs, writing job configs, and remembering to push checkpoints before the environment is destroyed, this skill handles submission, Hub authentication, timeout sizing, and Trackio monitoring as part of a single workflow.
When to use it
- Fine-tuning a Qwen or Llama model on a custom instruction dataset without owning a GPU
- Running DPO training on preference data to align a model's outputs
- Converting a freshly trained model to GGUF for use with Ollama or LM Studio
- Validating a dataset's column format before spending money on a GPU job
- Estimating training cost and time before committing to an a10g-large run