Low-Rank Adaptation: cheap fine-tuning that patches only small matrices.
1 min readupdated 2026-07-04
/ quick answer
LoRA freezes the base model and trains low-rank update matrices (a few % of parameters). Enables per-customer or per-task fine-tunes for cents instead of dollars. Low-Rank Adaptation: cheap fine-tuning that patches only small matrices.
Low-Rank Adaptation: cheap fine-tuning that patches only small matrices. LoRA freezes the base model and trains low-rank update matrices (a few % of parameters). Enables per-customer or per-task fine-tunes for cents instead of dollars. In practice: A 7B model gets 20 LoRA adapters, one per customer, all loaded dynamically. This dictionary node is part of the Onexial knowledge graph and links to related concepts, workflows and tools below.
Definition
LoRA freezes the base model and trains low-rank update matrices (a few % of parameters). Enables per-customer or per-task fine-tunes for cents instead of dollars.
Example
A 7B model gets 20 LoRA adapters, one per customer, all loaded dynamically.
/ frequently asked
What is LoRA?
LoRA freezes the base model and trains low-rank update matrices (a few % of parameters). Enables per-customer or per-task fine-tunes for cents instead of dollars.
What is an example of LoRA?
A 7B model gets 20 LoRA adapters, one per customer, all loaded dynamically.
Why does LoRA matter for AI and automation?
Low-Rank Adaptation: cheap fine-tuning that patches only small matrices. It connects to the workflows, prompts and tool stacks linked on this page, so you can move from definition to execution without leaving Onexial.