Models & ArchitectureUpdated 2026-10-24
Low-Rank Adaptation
Freezing the base model and training a small low-rank adapter for a new task.
#LoRA#微调#图像#适配
In plain language
LoRA cuts fine-tuning memory and cost, suiting style, domain and task adaptation. Multiple adapters can be switched or combined. It is less effective for injecting large amounts of new knowledge; use RAG for that.
Example
Train a LoRA on a few hundred product images to lock a brand visual style.
Recommended tools
Related terms
Models & ArchitectureFeatured
Diffusion Model
A generative model that turns random noise into images, video or audio through iterative denoising.
3 related terms
Engineering
Fine-tuning
Continue training a model on task data to better match a style or task.
3 related terms
Engineering
RAG vs Fine-tuning
RAG handles changing knowledge; fine-tuning handles behavior and style customization.
3 related terms
Applications
Agentic Workflow
Breaking a task into steps planned by AI, executed by tools and reviewed by humans.
3 related terms