LLM Finetuning plugin: what it installs and how to add it

The plugin structures an evaluation-first fine-tuning lifecycle for open-weight language and vision models. It helps select between LoRA or QLoRA SFT, preference optimisation with DPO, ORPO or KTO, GRPO or RLVR, and vision SFT, then prepares data, trains with Unsloth or TRL, evaluates checkpoints and exports quantised models.

Use it when you need to fine-tune a model end to end and want checkpoint promotion gated by evaluations. The workflow supports GGUF and FP8 export and includes trace-to-training-data guidance.

What's inside

• 10 skills: checkpoint-promotion, dataset-curation, eval-harness-first, finetuning-method-selection, grpo-rlvr-training, lora-qlora-recipes, preference-optimization, quantized-export, trace-to-training-data and vision-sft
• 2 slash commands: /finetune and /promote-checkpoint
• 3 agents: llm-finetuning-architect, llm-finetuning-eval-engineer and llm-finetuning-training-engineer