Work

Multi-Prompting Decoder Helps Better Language Understanding

Year: 2025

Type: conference-paper

Abstract: Recent large Pre-trained Language Models (PLMs) usually only provide users with the inference APIs, namely the emerging Modelas-a-Service (MaaS) setting.To adapt MaaS PLMs to downstream tasks without accessing their parameters and gradients, some existing methods focus on the output-side adaptation of PLMs, viewing the PLM as an encoder and then optimizing a task-specific decoder for decoding the output hidden states and class scores of the PLM.Despite the effectiveness of these methods, they only use a single prompt to query PLMs for decoding, leading to a heavy reliance on the quality of the adopted prompt.In this paper, we propose a simple yet effective Multi-Prompting Decoder (MPD) framework for MaaS adaptation.The core idea is to query PLMs with multiple different prompts for each sample, thereby obtaining multiple output hidden states and class scores from PLMs for subsequent decoding.Such multi-prompting decoding paradigm can simultaneously mitigate reliance on the quality of a single prompt, alleviate the issue of data scarcity under the few-shot setting, and provide richer knowledge extracted from PLMs.Specifically, we propose two decoding strategies: multi-prompting decoding with optimal transport for hidden states and calibrated decoding for class scores.Extensive experiments demonstrate that our method is effective on multiple natural language understanding datasets under the few-shot setting.

DOI: 10.18653/v1/2025.findings-acl.11