LLM fine-tuning on HF Jobs (Hugging Face)

A detailed guide for training models on Hugging Face Jobs. It covers SFT, DPO, GRPO and reward-model training with TRL or Unsloth, preparing and validating datasets, picking hardware and estimating cost, monitoring with Trackio, saving results to the Hub, and converting models to GGUF for local use. Bundled scripts and references support each step. Requires a Hugging Face account; GPU jobs need a paid plan.

Try asking:

• "Fine-tune Qwen on my support-ticket dataset with SFT"
• "Run DPO training on Hugging Face Jobs and push the model"
• "Convert my fine-tuned model to GGUF for llama.cpp"

Source: huggingface/skills · Apache-2.0 · by Hugging Face