PorTAL

Ramp Labs has open-sourced PorTAL, a framework for shared task representations and cross-model LoRA adaptation.

PorTAL learns base-model-agnostic task latents along with lightweight per-base alignment to generate generic layer-wise LoRA weights.

With this framework, a specific task only needs to be trained once, and it can then be instantly adapted to various supported frozen base models, with the resulting output exportable in the standard Hugging Face PEFT adapter format.