CombLM: Adapting Black-Box Language Models through Small Fine-Tuned Models
This addresses the challenge for practitioners who need to adapt high-quality black-box models efficiently, offering a lightweight solution with incremental improvements over existing adaptation methods.
The paper tackles the problem of adapting large black-box language models to new tasks and domains without access to their weights, by fine-tuning a small white-box model and combining it with the large model through a learned network, resulting in performance improvements of up to 9% while using a domain expert 23x smaller.
Methods for adapting language models (LMs) to new tasks and domains have traditionally assumed white-box access to the model, and work by modifying its parameters. However, this is incompatible with a recent trend in the field, where the highest quality models are only available as black-boxes through inference APIs. Even when the model weights are available, the computational cost of fine-tuning large LMs can be prohibitive for most practitioners. In this work, we present a lightweight method for adapting large LMs to new domains and tasks, assuming no access to their weights or intermediate activations. Our approach fine-tunes a small white-box LM and combines it with the large black-box LM at the probability level through a small network, learned on a small validation set. We validate our approach by adapting a large LM (OPT-30B) to several domains and a downstream task (machine translation), observing improved performance in all cases, of up to 9%, while using a domain expert 23x smaller.