AgentVerse: Facilitating Multi-Agent Collaboration and Exploring Emergent Behaviors in Agents

Autonomous agents empowered by Large Language Models (LLMs) have undergone significant improvements, enabling them to generalize across a broad spectrum of tasks. However, in real-world scenarios, cooperation among individuals is often required to enhance the efficiency and effectiveness of task accomplishment. Hence, inspired by human group dynamics, we propose a multi-agent framework \framework that can collaboratively and dynamically adjust its composition as a greater-than-the-sum-of-its-parts system. Our experiments demonstrate that \framework framework can effectively deploy multi-agent groups that outperform a single agent. Furthermore, we delve into the emergence of social behaviors among individual agents within a group during collaborative task accomplishment. In view of these behaviors, we discuss some possible strategies to leverage positive ones and mitigate negative ones for improving the collaborative potential of multi-agent groups. Our codes for \framework will soon be released at \url{https://github.com/OpenBMB/AgentVerse}.

ChatEval: Towards Better LLM-based Evaluators through Multi-Agent Debate

Text evaluation has historically posed significant challenges, often demanding substantial labor and time cost. With the emergence of large language models (LLMs), researchers have explored LLMs' potential as alternatives for human evaluation. While these single-agent-based approaches show promise, experimental results suggest that further advancements are needed to bridge the gap between their current effectiveness and human-level evaluation quality. Recognizing that best practices of human evaluation processes often involve multiple human annotators collaborating in the evaluation, we resort to a multi-agent debate framework, moving beyond single-agent prompting strategies. The multi-agent-based approach enables a group of LLMs to synergize with an array of intelligent counterparts, harnessing their distinct capabilities and expertise to enhance efficiency and effectiveness in handling intricate tasks. In this paper, we construct a multi-agent referee team called ChatEval to autonomously discuss and evaluate the quality of generated responses from different models on open-ended questions and traditional natural language generation (NLG) tasks. Our analysis shows that ChatEval transcends mere textual scoring, offering a human-mimicking evaluation process for reliable assessments. Our code is available at https://github.com/chanchimin/ChatEval.

Communicative Agents for Software Development

Software engineering is a domain characterized by intricate decision-making processes, often relying on nuanced intuition and consultation. Recent advancements in deep learning have started to revolutionize software engineering practices through elaborate designs implemented at various stages of software development. In this paper, we present an innovative paradigm that leverages large language models (LLMs) throughout the entire software development process, streamlining and unifying key processes through natural language communication, thereby eliminating the need for specialized models at each phase. At the core of this paradigm lies ChatDev, a virtual chat-powered software development company that mirrors the established waterfall model, meticulously dividing the development process into four distinct chronological stages: designing, coding, testing, and documenting. Each stage engages a team of agents, such as programmers, code reviewers, and test engineers, fostering collaborative dialogue and facilitating a seamless workflow. The chat chain acts as a facilitator, breaking down each stage into atomic subtasks. This enables dual roles, allowing for proposing and validating solutions through context-aware communication, leading to efficient resolution of specific subtasks. The instrumental analysis of ChatDev highlights its remarkable efficacy in software generation, enabling the completion of the entire software development process in under seven minutes at a cost of less than one dollar. It not only identifies and alleviates potential vulnerabilities but also rectifies potential hallucinations while maintaining commendable efficiency and cost-effectiveness. The potential of ChatDev unveils fresh possibilities for integrating LLMs into the realm of software development.

Arbitrary Few Parameters are Good Enough for Adapting Large-scale Pre-trained Language Models

Parameter-efficient tuning (PET) methods can effectively drive extremely large pre-trained language models (PLMs) by only training minimal parameters. Different PET methods utilize different manually designed modules. In a small PLM, there are usually noticeable performance differences among PET methods. Nevertheless, when a PLM's scale grows up to tens of billions of parameters, all PET methods achieve almost the same performance and even perform on par with the full-parameter fine-tuning method. Hence, we hypothesize that model scaling can mitigate the design differences (the module structures and the number of trainable parameters) among PET methods. To study this hypothesis, we introduce a more flexible PET method - arbitrary PET (APET) method - to be compatible with arbitrary module structures and any number of trainable parameters. Then, we experiment on $11$ NLP tasks of $5$ types and $2$ representative PLMs. From our investigations, we find that the model scaling (1) mitigates the effects of the arbitrary module structure on the performance of tuning methods, and (2) enables the tuning methods to optimize fewer parameters to achieve the full-parameter fine-tuning performance. Intriguingly, we also observe that all tuning methods require almost the same number of trainable parameters to drive PLMs. We discuss this phenomenon and the above two findings collectively from optimization perspectives to fathom the mechanisms behind them. These conclusions not only demonstrate the positive impact of model scaling on tuning methods but disclose its mechanisms, which help us design more effective and efficient tuning methods on larger-scale PLMs.

Tool Learning with Foundation Models

Humans possess an extraordinary ability to create and utilize tools, allowing them to overcome physical limitations and explore new frontiers. With the advent of foundation models, AI systems have the potential to be equally adept in tool use as humans. This paradigm, i.e., tool learning with foundation models, combines the strengths of specialized tools and foundation models to achieve enhanced accuracy, efficiency, and automation in problem-solving. Despite its immense potential, there is still a lack of a comprehensive understanding of key challenges, opportunities, and future endeavors in this field. To this end, we present a systematic investigation of tool learning in this paper. We first introduce the background of tool learning, including its cognitive origins, the paradigm shift of foundation models, and the complementary roles of tools and models. Then we recapitulate existing tool learning research into tool-augmented and tool-oriented learning. We formulate a general tool learning framework: starting from understanding the user instruction, models should learn to decompose a complex task into several subtasks, dynamically adjust their plan through reasoning, and effectively conquer each sub-task by selecting appropriate tools. We also discuss how to train models for improved tool-use capabilities and facilitate the generalization in tool learning. Considering the lack of a systematic tool learning evaluation in prior works, we experiment with 17 representative tools and show the potential of current foundation models in skillfully utilizing tools. Finally, we discuss several open problems that require further investigation for tool learning. Overall, we hope this paper could inspire future research in integrating tools with foundation models.

Human Emotion Knowledge Representation Emerges in Large Language Model and Supports Discrete Emotion Inference

How humans infer discrete emotions is a fundamental research question in the field of psychology. While conceptual knowledge about emotions (emotion knowledge) has been suggested to be essential for emotion inference, evidence to date is mostly indirect and inconclusive. As the large language models (LLMs) have been shown to support effective representations of various human conceptual knowledge, the present study further employed artificial neurons in LLMs to investigate the mechanism of human emotion inference. With artificial neurons activated by prompts, the LLM (RoBERTa) demonstrated a similar conceptual structure of 27 discrete emotions as that of human behaviors. Furthermore, the LLM-based conceptual structure revealed a human-like reliance on 14 underlying conceptual attributes of emotions for emotion inference. Most importantly, by manipulating attribute-specific neurons, we found that the corresponding LLM's emotion inference performance deteriorated, and the performance deterioration was correlated to the effectiveness of representations of the conceptual attributes on the human side. Our findings provide direct evidence for the emergence of emotion knowledge representation in large language models and suggest its casual support for discrete emotion inference.

Delta Tuning: A Comprehensive Study of Parameter Efficient Methods for Pre-trained Language Models

Despite the success, the process of fine-tuning large-scale PLMs brings prohibitive adaptation costs. In fact, fine-tuning all the parameters of a colossal model and retaining separate instances for different tasks are practically infeasible. This necessitates a new branch of research focusing on the parameter-efficient adaptation of PLMs, dubbed as delta tuning in this paper. In contrast with the standard fine-tuning, delta tuning only fine-tunes a small portion of the model parameters while keeping the rest untouched, largely reducing both the computation and storage costs. Recent studies have demonstrated that a series of delta tuning methods with distinct tuned parameter selection could achieve performance on a par with full-parameter fine-tuning, suggesting a new promising way of stimulating large-scale PLMs. In this paper, we first formally describe the problem of delta tuning and then comprehensively review recent delta tuning approaches. We also propose a unified categorization criterion that divide existing delta tuning methods into three groups: addition-based, specification-based, and reparameterization-based methods. Though initially proposed as an efficient method to steer large models, we believe that some of the fascinating evidence discovered along with delta tuning could help further reveal the mechanisms of PLMs and even deep neural networks. To this end, we discuss the theoretical principles underlying the effectiveness of delta tuning and propose frameworks to interpret delta tuning from the perspective of optimization and optimal control, respectively. Furthermore, we provide a holistic empirical study of representative methods, where results on over 100 NLP tasks demonstrate a comprehensive performance comparison of different approaches. The experimental results also cover the analysis of combinatorial, scaling and transferable properties of delta tuning.

On Transferability of Prompt Tuning for Natural Language Understanding

Prompt tuning (PT) is a promising parameter-efficient method to utilize extremely large pre-trained language models (PLMs), which could achieve comparable performance to full-parameter fine-tuning by only tuning a few soft prompts. However, compared to fine-tuning, PT empirically requires much more training steps. To explore whether we can improve the efficiency of PT by reusing trained soft prompts and sharing learned knowledge, we empirically investigate the transferability of soft prompts across different tasks and models. In cross-task transfer, we find that trained soft prompts can well transfer to similar tasks and initialize PT for them to accelerate training and improve performance. Moreover, to explore what factors influence prompts' transferability across tasks, we investigate how to measure the prompt similarity and find that the overlapping rate of activated neurons highly correlates to the transferability. In cross-model transfer, we explore how to project the prompts of a PLM to another PLM and successfully train a kind of projector which can achieve non-trivial transfer performance on similar tasks. However, initializing PT with the projected prompts does not work well, which may be caused by optimization preferences and PLMs' high redundancy. Our findings show that improving PT with knowledge transfer is possible and promising, while prompts' cross-task transferability is generally better than the cross-model transferability.

Exploring Low-dimensional Intrinsic Task Subspace via Prompt Tuning

How can pre-trained language models (PLMs) learn universal representations and effectively adapt to broad NLP tasks differing a lot superficially? In this work, we empirically find evidences indicating that the adaptations of PLMs to various tasks can be reparameterized as optimizing only a few free parameters in a common low-dimensional intrinsic task subspace, which may help us understand why PLMs could easily adapt to various NLP tasks with small-scale data. Specifically, to find such a subspace and examine its universality, we resort to the recent success of prompt tuning and decompose the soft prompts of multiple NLP tasks into the same low-dimensional nonlinear subspace, then we learn to adapt the PLM to unseen tasks or data by only tuning parameters in the subspace. We dub this pipeline as intrinsic prompt tuning (IPT). In experiments, we study diverse few-shot NLP tasks and surprisingly find that in a 5-dimensional subspace found with 100 random tasks, by only tuning 5 free parameters, we can recover 87% and 65% of the full prompt tuning performance for 100 seen tasks (using different training data) and 20 unseen tasks, respectively, showing great generalization ability of the found intrinsic task subspace. Besides being an analysis tool, IPT could further bring practical benefits, such as improving the prompt tuning stability.