Self-hosted alternatives to finetuning-method-selection

F

finetuning-method-selection

Decide whether to fine-tune at all, and route to the right method

finetuning-method-selection is an agent skill from the wshobson/agents collection that decides whether fine-tuning is the right tool for a model problem at all, and if it is, which method and which base-model size class to use. It is the router for the fine-tuning lifecycle. The other skills in that sequence assume this routing has already happened, so it is the place to start before working on training data, recipes or training runs. The methods it routes between include supervised fine-tuning (SFT), preference methods such as DPO, ORPO and KTO, reinforcement learning methods such as GRPO and RLVR, and continued pretraining. Each has a different job: SFT teaches a model to produce target responses, preference methods teach it which of two responses is better, and continued pretraining adds knowledge from large amounts of domain text. The skill helps the agent weigh those options against the problem, the data available and the size of the model that can be trained. Its job is to avoid the common mistake of reaching for fine-tuning when prompting, retrieval or a smaller change would do the job. The skill is installed with the skills CLI from the wshobson/agents repository, and its skills.sh page showed about 2.4K installs and passing results from three security audits at the time of the scan.