LLaMA-Factory Online

Fine-tuning an open-source language model usually means renting GPUs, wiring up training scripts and tuning parameters by hand.

LLaMA-Factory Online is a no-code LLM fine-tuning platform that moves that work into a visual interface, built in official collaboration with the LLaMA-Factory open-source project.

You set parameters through the UI instead of writing code, and the platform covers the workflow from data preparation to chatting with the finished model.

It supports over 100 mainstream open-source models and datasets.

Training methods include SFT, PPO, DPO and KTO, with 16bit, LoRA and QLoRA precisions, plus optimization algorithms such as GaLore, BAdam and PiSSA.

Distributed training is supported, and GPU resources, including H800A cards, are elastic and billed by the second, which suits short experiments.

New users get vouchers and promotional discounts.

Customer support and a community are available if you get stuck.

The site is in Chinese, so check that before you plan around it.