Deploying Multi-task Online Server with Large Language Model

TL;DR

This work presents a three-stage multi-task learning framework for large language models that involves task filtering, followed by fine-tuning on high-resource tasks, and finally fine-tuning on all tasks.

Abstract

In the industry, numerous tasks are deployed online. Traditional approaches often tackle each task separately by its own network, which leads to excessive costs for developing and scaling models, especially in the context of large language models. Although multi-task methods can save costs through parameter sharing, they often struggle to outperform single-task methods in real-world applications. To tackle these challenges, we present a three-stage multi-task learning framework for large language models. It involves task filtering, followed by fine-tuning on high-resource tasks, and finally fine-tuning on all tasks. We conducted comprehensive experiments in single-task and multi-task settings. Our approach, exemplified on different benchmarks, demonstrates that it is able to achieve performance comparable to the single-task method while reducing up to 90.9\% of its overhead.

Figures (3)

• Figure 1: Two types of online serving strategies. (a) Independent single-task models are trained and deployed for each task. (b) One multi-task model is trained and deployed for all tasks.
• Figure 2: Pipeline of the proposed method. It starts with domain-specific continual pre-training, where the model undergoes self-supervised learning using domain-specific data. Next, we perform multi-task fine-tuning on high-resource tasks. Then, we perform multi-task fine-tuning on all tasks, enabling the model to learn from a mixture of tasks simultaneously. Finally, the multi-task model is deployed online to serve different tasks.
• Figure 3: Pipeline of data processing.