Definition
Plain language
To stop a model's weights from changing during further training.
As stated in the literature
To hold a set of parameters fixed during optimization, typically while training only adapters, communication layers, or downstream modules.
Also called: frozen, frozen weights
Why it matters: Freezing the bulk of a model keeps training cheap and protects pretrained knowledge from being overwritten by a small task-specific dataset.
For example, when adding a small LoRA adapter to a 70B base model, you freeze the original weights and only train the few million adapter parameters.
Heard on the show
“That's the difference between an agent that feels alive and one that's effectively frozen.”Episode 016 — Why Your Coding Agent Stalls While the GPU Runs Hot