Branch-Solve-Merge Improves Large Language Model Evaluation and Generation
This addresses the problem of improving LLM performance on complex natural language tasks for users needing more reliable evaluation and generation, though it appears incremental as it builds on existing LLM program methods.
The paper tackles the problem of LLMs' lack of coherence and inability to plan for multi-faceted tasks by proposing Branch-Solve-Merge (BSM), a program that decomposes tasks into sub-tasks, solves them in parallel, and merges solutions. The result shows improvements in evaluation correctness and consistency, with up to 26% better human-LLM agreement and up to 50% reduced biases, and in constrained text generation, with 12% better constraint satisfaction.
Large Language Models (LLMs) are frequently used for multi-faceted language generation and evaluation tasks that involve satisfying intricate user constraints or taking into account multiple aspects and criteria. However, their performance can fall short, due to the model's lack of coherence and inability to plan and decompose the problem. We propose Branch-Solve-Merge (BSM), a Large Language Model program (Schlag et al., 2023) for tackling such challenging natural language tasks. It consists of branch, solve, and merge modules that are parameterized with specific prompts to the base LLM. These three modules plan a decomposition of the task into multiple parallel sub-tasks, independently solve them, and fuse the solutions to the sub-tasks. We apply our method to the tasks of LLM response evaluation and constrained text generation and evaluate its effectiveness with multiple LLMs, including Vicuna, LLaMA-2-chat, and GPT-4. BSM improves the evaluation correctness and consistency for each LLM by enhancing human-LLM agreement by up to 26%, reducing length and pairwise position biases by up to 50%, and allowing LLaMA2-chat to match or outperform GPT-4 on most domains. On a constraint story generation task, BSM improves the coherence of stories while also improving constraint satisfaction by 12%.