LLM Reasoners: New Evaluation, Library, and Analysis of Step-by-Step Reasoning with Large Language Models

Generating accurate step-by-step reasoning is essential for Large Language Models (LLMs) to address complex problems and enhance robustness and interpretability. Despite the flux of research on developing advanced reasoning approaches, systematically analyzing the diverse LLMs and reasoning strategies in generating reasoning chains remains a significant challenge. The difficulties stem from the lack of two key elements: (1) an automatic method for evaluating the generated reasoning chains on different tasks, and (2) a unified formalism and implementation of the diverse reasoning approaches for systematic comparison. This paper aims to close the gap: (1) We introduce AutoRace for fully automated reasoning chain evaluation. Existing metrics rely on expensive human annotations or pre-defined LLM prompts not adaptable to different tasks. In contrast, AutoRace automatically creates detailed evaluation criteria tailored for each task, and uses GPT-4 for accurate evaluation following the criteria. (2) We develop LLM Reasoners, a library for standardized modular implementation of existing and new reasoning algorithms, under a unified formulation of the search, reward, and world model components. With the new evaluation and library, (3) we conduct extensive study of different reasoning approaches (e.g., CoT, ToT, RAP). The analysis reveals interesting findings about different factors contributing to reasoning, including the reward-guidance, breadth-vs-depth in search, world model, and prompt formats, etc.

Reasoning with Language Model is Planning with World Model

Large language models (LLMs) have shown remarkable reasoning capabilities, especially when prompted to generate intermediate reasoning steps (e.g., Chain-of-Thought, CoT). However, LLMs can still struggle with problems that are easy for humans, such as generating action plans for executing tasks in a given environment, or performing complex math, logical, and commonsense reasoning. The deficiency stems from the key fact that LLMs lack an internal $\textit{world model}$ to predict the world $\textit{state}$ (e.g., environment status, intermediate variable values) and simulate long-term outcomes of actions. This prevents LLMs from performing deliberate planning akin to human brains, which involves exploring alternative reasoning paths, anticipating future states and rewards, and iteratively refining existing reasoning steps. To overcome the limitations, we propose a new LLM reasoning framework, $\underline{R}\textit{easoning vi}\underline{a} \underline{P}\textit{lanning}$ $\textbf{(RAP)}$. RAP repurposes the LLM as both a world model and a reasoning agent, and incorporates a principled planning algorithm (based on Monto Carlo Tree Search) for strategic exploration in the vast reasoning space. During reasoning, the LLM (as agent) incrementally builds a reasoning tree under the guidance of the LLM (as world model) and task-specific rewards, and obtains a high-reward reasoning path efficiently with a proper balance between exploration $\textit{vs.}$ exploitation. We apply RAP to a variety of challenging reasoning problems including plan generation, math reasoning, and logical inference. Empirical results on these tasks demonstrate the superiority of RAP over various strong baselines, including CoT and least-to-most prompting with self-consistency. RAP on LLAMA-33B surpasses CoT on GPT-4 with 33% relative improvement in a plan generation setting.

ToolkenGPT: Augmenting Frozen Language Models with Massive Tools via Tool Embeddings

Augmenting large language models (LLMs) with external tools has emerged as a promising approach to solving complex problems. However, traditional methods, which finetune LLMs with tool demonstration data, can be both costly and restricted to a predefined set of tools. Recent in-context learning paradigm alleviates these issues, but the limited context length only allows for a few shots of demonstrations, leading to suboptimal understandings of the tools. Moreover, when there are numerous tools to choose from, in-context learning could completely fail to work. In this paper, we propose an alternative approach, $\textbf{ToolkenGPT}$, which combines the benefits of both sides. Our approach represents each $\underline{tool}$ as a to$\underline{ken}$ ($\textit{toolken}$) and learns an embedding for it, enabling tool calls in the same way as generating a regular word token. Once a toolken is triggered, the LLM is prompted to complete arguments for the tool to execute. ToolkenGPT offers the flexibility to plug in an arbitrary number of tools by expanding the set of toolkens on the fly. In addition, it improves tool use by allowing extensive demonstration data for learning the toolken embeddings. In diverse domains, including numerical reasoning, knowledge-based question answering, and embodied plan generation, our approach effectively augments LLMs with tools and substantially outperforms various latest baselines. ToolkenGPT demonstrates the promising ability to use relevant tools from a large tool set in complex scenarios.

BertNet: Harvesting Knowledge Graphs from Pretrained Language Models

Symbolic knowledge graphs (KGs) have been constructed either by expensive human crowdsourcing or with domain-specific complex information extraction pipelines. The emerging large pretrained language models (LMs), such as Bert, have shown to implicitly encode massive knowledge which can be queried with properly designed prompts. However, compared to the explicit KGs, the implict knowledge in the black-box LMs is often difficult to access or edit and lacks explainability. In this work, we aim at harvesting symbolic KGs from the LMs, a new framework for automatic KG construction empowered by the neural LMs' flexibility and scalability. Compared to prior works that often rely on large human annotated data or existing massive KGs, our approach requires only the minimal definition of relations as inputs, and hence is suitable for extracting knowledge of rich new relations not available before.The approach automatically generates diverse prompts, and performs efficient knowledge search within a given LM for consistent and extensive outputs. The harvested knowledge with our approach is substantially more accurate than with previous methods, as shown in both automatic and human evaluation. As a result, we derive from diverse LMs a family of new KGs (e.g., BertNet and RoBERTaNet) that contain a richer set of commonsense relations, including complex ones (e.g., "A is capable of but not good at B"), than the human-annotated KGs (e.g., ConceptNet). Besides, the resulting KGs also serve as a vehicle to interpret the respective source LMs, leading to new insights into the varying knowledge capability of different LMs.

Does Recommend-Revise Produce Reliable Annotations? An Analysis on Missing Instances in DocRED

DocRED is a widely used dataset for document-level relation extraction. In the large-scale annotation, a \textit{recommend-revise} scheme is adopted to reduce the workload. Within this scheme, annotators are provided with candidate relation instances from distant supervision, and they then manually supplement and remove relational facts based on the recommendations. However, when comparing DocRED with a subset relabeled from scratch, we find that this scheme results in a considerable amount of false negative samples and an obvious bias towards popular entities and relations. Furthermore, we observe that the models trained on DocRED have low recall on our relabeled dataset and inherit the same bias in the training data. Through the analysis of annotators' behaviors, we figure out the underlying reason for the problems above: the scheme actually discourages annotators from supplementing adequate instances in the revision phase. We appeal to future research to take into consideration the issues with the recommend-revise scheme when designing new models and annotation schemes. The relabeled dataset is released at \url{https://github.com/AndrewZhe/Revisit-DocRED}, to serve as a more reliable test set of document RE models.

Benchmarking Commonsense Knowledge Base Population with an Effective Evaluation Dataset

Reasoning over commonsense knowledge bases (CSKB) whose elements are in the form of free-text is an important yet hard task in NLP. While CSKB completion only fills the missing links within the domain of the CSKB, CSKB population is alternatively proposed with the goal of reasoning unseen assertions from external resources. In this task, CSKBs are grounded to a large-scale eventuality (activity, state, and event) graph to discriminate whether novel triples from the eventuality graph are plausible or not. However, existing evaluations on the population task are either not accurate (automatic evaluation with randomly sampled negative examples) or of small scale (human annotation). In this paper, we benchmark the CSKB population task with a new large-scale dataset by first aligning four popular CSKBs, and then presenting a high-quality human-annotated evaluation set to probe neural models' commonsense reasoning ability. We also propose a novel inductive commonsense reasoning model that reasons over graphs. Experimental results show that generalizing commonsense reasoning on unseen assertions is inherently a hard task. Models achieving high accuracy during training perform poorly on the evaluation set, with a large gap between human performance. We will make the data publicly available for future contributions. Codes and data are available at https://github.com/HKUST-KnowComp/CSKB-Population.