Introduction

LLaMA Factory is an open-source low-code framework for LLM fine-tuning. It bundles the most widely-used tuning techniques and ships a zero-code WebUI, making it one of the most popular tuning frameworks in the open-source community.

Alaya NeW Cloud is deeply integrated with LLaMA Factory. This section covers the full path of running LLaMA Factory on Alaya NeW — concepts, single- and multi-node experiments, and storage selection trade-offs.

Use cases

LLaMA Factory's lightweight, modular design substantially lowers the cost of adapting large models to complex scenarios. Common applications:

• Domain-specific fine-tuning — medical, legal, financial, cultural multimodal LLM tuning
• Task-specific optimization — text generation, classification, QA, translation
• Resource-constrained scenarios — LoRA / QLoRA fine-tuning on memory-limited GPUs
• Multimodal training — combine text + image + audio data for multimodal-input models
• Rapid customization — quick path for AI engineers, researchers, and enterprise teams to ship custom LLMs

Highlights

LLaMA Factory is open-sourced by Beihang University and purpose-built for LLM fine-tuning. Key capabilities:

• Efficient and low-cost — supports 100+ models with a streamlined fine-tuning pipeline
• Zero-code WebUI — model selection, dataset prep, training, evaluation, export — no code required
• Rich dataset options — built-in datasets plus custom Alpaca / ShareGPT formats
• Diverse algorithms — LoRA, GaLore, DoRA, and more
• Live monitoring — TensorBoard, WanDB, MLflow, SwanLab integrations
• Fast inference — vLLM-backed OpenAI-style API, browser UI, and CLI

Section map