Adapter
A small set of trainable parameters inserted into a pretrained model. During fine-tuning, the backbone weights are frozen and only the adapter is updated. It belongs to parameter-efficient fine-tuning (PEFT) methods, with low training cost and little disruption to the original model's capabilities.
Timeline (last 90 days)
Although the Adapter method reduces training parameters, it introduces extra layers at inference time, increases latency, and cannot be eliminated through matrix merging.
Rubric-CEPR distills the validated best candidate back into the editor through a lightweight adapter, updating only a small number of parameters rather than retraining the entire model.
A drawback of the Adapter method is increased inference latency, because information flow must pass serially through the additionally inserted modules.
All Facts (3)
Although the Adapter method reduces training parameters, it introduces extra layers at inference time, increases latency, and cannot be eliminated through matrix merging.
50%UnverifiedRubric-CEPR distills the validated best candidate back into the editor through a lightweight adapter, updating only a small number of parameters rather than retraining the entire model.
50%UnverifiedA drawback of the Adapter method is increased inference latency, because information flow must pass serially through the additionally inserted modules.
50%