GPT vs RETRO: Exploring the Intersection of Retrieval and Parameter-Efficient Fine-Tuning
This work provides a comparative analysis for researchers and practitioners adapting large language models efficiently, but it is incremental as it applies existing methods to new models.
The paper compared Parameter-Efficient Fine-Tuning (PEFT) methods applied to GPT and Retrieval-Enhanced Transformer (RETRO) models, finding that RETRO outperforms GPT in zero-shot settings, but GPT has higher potential with PEFT, with 8B models offering an optimal cost-performance balance and P-tuning lagging behind other techniques.
Parameter-Efficient Fine-Tuning (PEFT) and Retrieval-Augmented Generation (RAG) have become popular methods for adapting large language models while minimizing compute requirements. In this paper, we apply PEFT methods (P-tuning, Adapters, and LoRA) to a modified Retrieval-Enhanced Transformer (RETRO) and a baseline GPT model across several sizes, ranging from 823 million to 48 billion parameters. We show that RETRO models outperform GPT models in zero-shot settings due to their unique pre-training process but GPT models have higher performance potential with PEFT. Additionally, our study indicates that 8B parameter models strike an optimal balance between cost and performance and P-tuning lags behind other PEFT techniques. We further provide a comparative analysis between applying PEFT to an Instruction-tuned RETRO model and base RETRO model. This work presents the first comprehensive comparison of various PEFT methods integrated with RAG, applied to both GPT and RETRO models, highlighting their relative performance.