Instant Transformer Adaption via HyperLoRA

Rujikorn Charakorn, Edoardo Cetin, Yujin Tang, Robert Tjarko Lange

Keywords: hypernetworks, finetuning, language models, zero-shot generalization

TL;DR: HyperLoRA generates LoRAs for domain-specific tasks given a language task description in a single forward pass

Abstract:

While Foundation Models provide a general tool for rapid content creation, they regularly require task-specific adaptation. Traditionally, this exercise involves careful curation of datasets and repeated fine-tuning of the underlying model. Fine-tuning techniques enable practitioners to adapt foundation models for many new applications but require expensive and lengthy training while being notably sensitive to hyper-parameter choices. To overcome these limitations, we introduce HyperLoRA, a model capable of adapting Large Language Models on the fly---solely based on a natural language description of the target task. HyperLoRA is a hypernetwork trained to construct LoRAs in a single inexpensive forward pass. After training HyperLoRA, we show that the ad-hoc reconstructed LoRA instances match the performance of task-specific adapters across the corresponding test sets.
Furthermore, HyperLoRA can compress hundreds of LoRAs instances and zero-shot generalize to entirely unseen tasks. This approach provides a significant step towards democratizing the specialization of foundation models and enables language-based adaptation with minimal compute requirements.

Submission Number: 64