本文へスキップ
Kitbay
スキル公式

huggingface-llm-trainer

Apache-2.02026年11月11日 更新11,334GitHubスタースクリプトあり

Hugging Face's skill for fine-tuning language and vision-language models on Hugging Face Jobs, its managed cloud GPUs, using TRL or Unsloth.

スキルの内容

Hugging Face's skill for fine-tuning language and vision-language models on Hugging Face Jobs, its managed cloud GPUs, using TRL or Unsloth. It walks the agent through choosing a method (supervised fine-tuning, DPO, GRPO or reward modelling), preparing and validating the dataset, writing a self-contained uv script, picking hardware and estimating cost, submitting and monitoring the job, and pushing the trained model to the Hub. It also covers converting results to GGUF for local use. Bundled scripts and reference notes support each step. A Hugging Face account with Jobs access and a write token are required, and jobs are billed.

できること

  • Fine-tune models with SFT, DPO, GRPO or reward modelling
  • Submit and monitor training runs on Hugging Face Jobs
  • Validate datasets and estimate hardware and cost
  • Convert trained models to GGUF