PyTorchKR Papers

“Code Generation”

총 10,163편
태그·기간 등 상세 필터

카테고리 탐색

arXiv:2608.29588v1cs.AI

스스로 이웃을 호출하기: 목적지 조건부 온-정책 자기 증류를 활용한 그래프 순회

Call Neighbours Yourself: Graph Walks with Destination-Conditioned On-Policy Self-Distillation

텍스트 속성 그래프(TAG)에 대한 추론은 대규모 언어 모델(LLM)이 노드의 텍스트와 이웃 전반에 분산된 근거를 결합할 것을 요구합니다. 기존 방법들은 생성 전에 접근 가능한 이웃 집합을 고정하므로 추론이 정적인 컨텍스트에서 수행되도록 강제하며, 모델이 추론 중에 누락된 근거를 획득하는 것을 방해합니다. 우리는 이웃 선택 자체가 추론 과정의 일부여야 한다고 주장합니다. 이를 위해, 우리는 LLM이 위상 제약 그래프 순회(graph-walk) 행동을 통해 그래프 이웃을 능동적으로 탐험할 수 있도록 지원하는 프레임워크인 Call Neighbours Yourself(CNY)를 제안합니다. CNY는 미리 선택된 이웃 집합에 대해 추론하는 대신 경량화된 이웃 미리보기를 노출하고 추가적인 근거를 위해 후보 이웃을 언제 확장할지 학습합니다. 이웃 탐험의 지연된 보상(delayed-credit) 문제를 해결하기 위해, 우리는 목적지 조건부 온-정책 자기 증류(destination-conditioned on-policy self-distillation)를 도입합니다. 이는 이웃의 콘텐츠가 공개된 후 선택된 이웃을 소급하여 평가하고, 그 결과로 나타나는 행동 선호도의 변화를 행동 수준의 학습 신호로 변환합니다. 통일된 원시 텍스트 설정 하에서 표준 TAG 추론 벤치마크에 대해 수행한 실험은 CNY가 고정 컨텍스트 사후 학습 베이스라인들을 일관되게 능가함을 보여줍니다. 나아가, 학습된 탐험 정책은 학습 중에 접하지 않은 새로운 그래프 및 그래프 수준 작업으로도 전이됩니다. 코드는 https://github.com/superallen13/CNY 에서 확인할 수 있습니다.

arXiv:2610.16135v1cs.CL

옴니모달 LLM은 말하는 것보다 더 많이 알고 있다

Omni-Modal LLMs Know More Than They Say

옴니모달 대규모 언어 모델(Omni-LLM)은 언어와 함께 시각, 오디오, IMU 신호 등 이질적인 모달리티(modality)를 처리하도록 LLM을 확장하여 더 폭넓은 멀티모달 과제를 가능하게 한다. 그러나 우리는 이러한 모델들이 비텍스트 모달리티의 정보를 정확하게 인지하고도 이를 제대로 표현하지 못하는 경우가 많다는 것을 발견했다. 우리는 이러한 단절 현상을 'knows-but-cannot-say(알지만 말할 수 없음)'라 명명한다. 이 현상을 조사하기 위해 우리는 개별 모달리티 표현과 응답 표현 간의 정렬을 측정하는 토큰 수준 모달리티 정렬(Token-level Modality Alignment, TMA)을 도입한다. 우리의 분석에 따르면 비텍스트 표현은 텍스트 표현에 비해 응답 토큰과의 정렬 정도가 낮고 받는 어텐션도 적은 것으로 나타났으며, 이는 모달리티 선호가 불균형하게 나타나는 원인을 설명해줄 수 있다. 이러한 발견에 착안하여 우리는 원본 모달리티 입력을 스스로 생성한 텍스트로 보완하여 비텍스트 지식에 더 쉽게 접근할 수 있게 하는 간단한 지각(perceive) → 추론(infer) 전략인 OmniCatalyst를 제안한다. 7개의 Omni-LLM 전반에서 OmniCatalyst는 일관되게 성능을 향상시키며(+7.29%) 더 균형 잡힌 모달리티 선호를 만들어내고, 다양한 학습 전략을 통해 추가적인 성능 향상도 얻을 수 있다. 추가 분석을 통해 지식 추출 단계가 TMA를 높인다는 것을 보였으며, 이는 정렬이 향상되면 응답 생성 과정에서 비텍스트 정보에 더 쉽게 접근할 수 있게 됨을 시사한다. 요약하면, 우리의 연구 결과는 Omni-LLM이 언어를 통해 모달리티 지식을 더 잘 표현할 수 있는 방법을 밝혀준다. 코드는 https://github.com/naver-ai/OmniLLM 에서 확인할 수 있다.

arXiv:1804.09399공식 코드 · Papers with Code

Convolutional Generative Adversarial Networks with Binary Neurons for Polyphonic Music Generation

It has been shown recently that deep convolutional generative adversarial networks (GANs) can learn to generate music in the form of piano-rolls, which represent music by binary-valued time-pitch matrices. However, existing models can only generate real-valued piano-rolls and require further post-processing, such as hard thresholding (HT) or Bernoulli sampling (BS), to obtain the final binary-valued results. In this paper, we study whether we can have a convolutional GAN model that directly creates binary-valued piano-rolls by using binary neurons. Specifically, we propose to append to the generator an additional refiner network, which uses binary neurons at the output layer. The whole network is trained in two stages. Firstly, the generator and the discriminator are pretrained. Then, the refiner network is trained along with the discriminator to learn to binarize the real-valued piano-rolls the pretrained generator creates. Experimental results show that using binary neurons instead of HT or BS indeed leads to better results in a number of objective measures. Moreover, deterministic binary neurons perform better than stochastic ones in both objective measures and a subjective test. The source code, training data and audio examples of the generated results can be found at https://salu133445.github.io/bmusegan/ .

arXiv:1904.09229공식 코드 · Papers with Code

XLSor: A Robust and Accurate Lung Segmentor on Chest X-Rays Using Criss-Cross Attention and Customized Radiorealistic Abnormalities Generation

This paper proposes a novel framework for lung segmentation in chest X-rays. It consists of two key contributions, a criss-cross attention based segmentation network and radiorealistic chest X-ray image synthesis (i.e. a synthesized radiograph that appears anatomically realistic) for data augmentation. The criss-cross attention modules capture rich global contextual information in both horizontal and vertical directions for all the pixels thus facilitating accurate lung segmentation. To reduce the manual annotation burden and to train a robust lung segmentor that can be adapted to pathological lungs with hazy lung boundaries, an image-to-image translation module is employed to synthesize radiorealistic abnormal CXRs from the source of normal ones for data augmentation. The lung masks of synthetic abnormal CXRs are propagated from the segmentation results of their normal counterparts, and then serve as pseudo masks for robust segmentor training. In addition, we annotate 100 CXRs with lung masks on a more challenging NIH Chest X-ray dataset containing both posterioranterior and anteroposterior views for evaluation. Extensive experiments validate the robustness and effectiveness of the proposed framework. The code and data can be found from https://github.com/rsummers11/CADLab/tree/master/Lung_Segmentation_XLSor .

arXiv:2003.14034공식 코드 · Papers with Code

SPARE3D: A Dataset for SPAtial REasoning on Three-View Line Drawings

Spatial reasoning is an important component of human intelligence. We can imagine the shapes of 3D objects and reason about their spatial relations by merely looking at their three-view line drawings in 2D, with different levels of competence. Can deep networks be trained to perform spatial reasoning tasks? How can we measure their "spatial intelligence"? To answer these questions, we present the SPARE3D dataset. Based on cognitive science and psychometrics, SPARE3D contains three types of 2D-3D reasoning tasks on view consistency, camera pose, and shape generation, with increasing difficulty. We then design a method to automatically generate a large number of challenging questions with ground truth answers for each task. They are used to provide supervision for training our baseline models using state-of-the-art architectures like ResNet. Our experiments show that although convolutional networks have achieved superhuman performance in many visual learning tasks, their spatial reasoning performance on SPARE3D tasks is either lower than average human performance or even close to random guesses. We hope SPARE3D can stimulate new problem formulations and network designs for spatial reasoning to empower intelligent robots to operate effectively in the 3D world via 2D sensors. The dataset and code are available at https://ai4ce.github.io/SPARE3D.

arXiv:2006.06762

Ansor: Generating High-Performance Tensor Programs for Deep Learning

High-performance tensor programs are crucial to guarantee efficient execution of deep neural networks. However, obtaining performant tensor programs for different operators on various hardware platforms is notoriously challenging. Currently, deep learning systems rely on vendor-provided kernel libraries or various search strategies to get performant tensor programs. These approaches either require significant engineering effort to develop platform-specific optimization code or fall short of finding high-performance programs due to restricted search space and ineffective exploration strategy. We present Ansor, a tensor program generation framework for deep learning applications. Compared with existing search strategies, Ansor explores many more optimization combinations by sampling programs from a hierarchical representation of the search space. Ansor then fine-tunes the sampled programs with evolutionary search and a learned cost model to identify the best programs. Ansor can find high-performance programs that are outside the search space of existing state-of-the-art approaches. In addition, Ansor utilizes a task scheduler to simultaneously optimize multiple subgraphs in deep neural networks. We show that Ansor improves the execution performance of deep neural networks relative to the state-of-the-art on the Intel CPU, ARM CPU, and NVIDIA GPU by up to $3.8\times$, $2.6\times$, and $1.7\times$, respectively.

arXiv:2006.13026공식 코드 · Papers with Code

Deep Polynomial Neural Networks

Deep Convolutional Neural Networks (DCNNs) are currently the method of choice both for generative, as well as for discriminative learning in computer vision and machine learning. The success of DCNNs can be attributed to the careful selection of their building blocks (e.g., residual blocks, rectifiers, sophisticated normalization schemes, to mention but a few). In this paper, we propose $\Pi$-Nets, a new class of function approximators based on polynomial expansions. $\Pi$-Nets are polynomial neural networks, i.e., the output is a high-order polynomial of the input. The unknown parameters, which are naturally represented by high-order tensors, are estimated through a collective tensor factorization with factors sharing. We introduce three tensor decompositions that significantly reduce the number of parameters and show how they can be efficiently implemented by hierarchical neural networks. We empirically demonstrate that $\Pi$-Nets are very expressive and they even produce good results without the use of non-linear activation functions in a large battery of tasks and signals, i.e., images, graphs, and audio. When used in conjunction with activation functions, $\Pi$-Nets produce state-of-the-art results in three challenging tasks, i.e. image generation, face verification and 3D mesh representation learning. The source code is available at \url{https://github.com/grigorisg9gr/polynomial_nets}.

arXiv:2007.09267공식 코드 · Papers with Code

Meshing Point Clouds with Predicted Intrinsic-Extrinsic Ratio Guidance

We are interested in reconstructing the mesh representation of object surfaces from point clouds. Surface reconstruction is a prerequisite for downstream applications such as rendering, collision avoidance for planning, animation, etc. However, the task is challenging if the input point cloud has a low resolution, which is common in real-world scenarios (e.g., from LiDAR or Kinect sensors). Existing learning-based mesh generative methods mostly predict the surface by first building a shape embedding that is at the whole object level, a design that causes issues in generating fine-grained details and generalizing to unseen categories. Instead, we propose to leverage the input point cloud as much as possible, by only adding connectivity information to existing points. Particularly, we predict which triplets of points should form faces. Our key innovation is a surrogate of local connectivity, calculated by comparing the intrinsic/extrinsic metrics. We learn to predict this surrogate using a deep point cloud network and then feed it to an efficient post-processing module for high-quality mesh generation. We demonstrate that our method can not only preserve details, handle ambiguous structures, but also possess strong generalizability to unseen categories by experiments on synthetic and real data. The code is available at https://github.com/Colin97/Point2Mesh.

arXiv:2009.08119공식 코드 · Papers with Code

Collaborative Training between Region Proposal Localization and Classification for Domain Adaptive Object Detection

Object detectors are usually trained with large amount of labeled data, which is expensive and labor-intensive. Pre-trained detectors applied to unlabeled dataset always suffer from the difference of dataset distribution, also called domain shift. Domain adaptation for object detection tries to adapt the detector from labeled datasets to unlabeled ones for better performance. In this paper, we are the first to reveal that the region proposal network (RPN) and region proposal classifier~(RPC) in the endemic two-stage detectors (e.g., Faster RCNN) demonstrate significantly different transferability when facing large domain gap. The region classifier shows preferable performance but is limited without RPN's high-quality proposals while simple alignment in the backbone network is not effective enough for RPN adaptation. We delve into the consistency and the difference of RPN and RPC, treat them individually and leverage high-confidence output of one as mutual guidance to train the other. Moreover, the samples with low-confidence are used for discrepancy calculation between RPN and RPC and minimax optimization. Extensive experimental results on various scenarios have demonstrated the effectiveness of our proposed method in both domain-adaptive region proposal generation and object detection. Code is available at https://github.com/GanlongZhao/CST_DA_detection.

arXiv:2012.12306

Pattern Recognition Scheme for Large-Scale Cloud Detection over Landmarks

Landmark recognition and matching is a critical step in many Image Navigation and Registration (INR) models for geostationary satellite services, as well as to maintain the geometric quality assessment (GQA) in the instrument data processing chain of Earth observation satellites. Matching the landmark accurately is of paramount relevance, and the process can be strongly impacted by the cloud contamination of a given landmark. This paper introduces a complete pattern recognition methodology able to detect the presence of clouds over landmarks using Meteosat Second Generation (MSG) data. The methodology is based on the ensemble combination of dedicated support vector machines (SVMs) dependent on the particular landmark and illumination conditions. This divide-and-conquer strategy is motivated by the data complexity and follows a physically-based strategy that considers variability both in seasonality and illumination conditions along the day to split observations. In addition, it allows training the classification scheme with millions of samples at an affordable computational costs. The image archive was composed of 200 landmark test sites with near 7 million multispectral images that correspond to MSG acquisitions during 2010. Results are analyzed in terms of cloud detection accuracy and computational cost. We provide illustrative source code and a portion of the huge training data to the community.

arXiv:2109.12674공식 코드 · Papers with Code

MetaDrive: Composing Diverse Driving Scenarios for Generalizable Reinforcement Learning

Driving safely requires multiple capabilities from human and intelligent agents, such as the generalizability to unseen environments, the safety awareness of the surrounding traffic, and the decision-making in complex multi-agent settings. Despite the great success of Reinforcement Learning (RL), most of the RL research works investigate each capability separately due to the lack of integrated environments. In this work, we develop a new driving simulation platform called MetaDrive to support the research of generalizable reinforcement learning algorithms for machine autonomy. MetaDrive is highly compositional, which can generate an infinite number of diverse driving scenarios from both the procedural generation and the real data importing. Based on MetaDrive, we construct a variety of RL tasks and baselines in both single-agent and multi-agent settings, including benchmarking generalizability across unseen scenes, safe exploration, and learning multi-agent traffic. The generalization experiments conducted on both procedurally generated scenarios and real-world scenarios show that increasing the diversity and the size of the training set leads to the improvement of the RL agent's generalizability. We further evaluate various safe reinforcement learning and multi-agent reinforcement learning algorithms in MetaDrive environments and provide the benchmarks. Source code, documentation, and demo video are available at \url{ https://metadriverse.github.io/metadrive}.

arXiv:2203.02250공식 코드 · Papers with Code

Patch Similarity Aware Data-Free Quantization for Vision Transformers

Vision transformers have recently gained great success on various computer vision tasks; nevertheless, their high model complexity makes it challenging to deploy on resource-constrained devices. Quantization is an effective approach to reduce model complexity, and data-free quantization, which can address data privacy and security concerns during model deployment, has received widespread interest. Unfortunately, all existing methods, such as BN regularization, were designed for convolutional neural networks and cannot be applied to vision transformers with significantly different model architectures. In this paper, we propose PSAQ-ViT, a Patch Similarity Aware data-free Quantization framework for Vision Transformers, to enable the generation of "realistic" samples based on the vision transformer's unique properties for calibrating the quantization parameters. Specifically, we analyze the self-attention module's properties and reveal a general difference (patch similarity) in its processing of Gaussian noise and real images. The above insights guide us to design a relative value metric to optimize the Gaussian noise to approximate the real images, which are then utilized to calibrate the quantization parameters. Extensive experiments and ablation studies are conducted on various benchmarks to validate the effectiveness of PSAQ-ViT, which can even outperform the real-data-driven methods. Code is available at: https://github.com/zkkli/PSAQ-ViT.

arXiv:2204.03655

Learning to Walk Autonomously via Reset-Free Quality-Diversity

Quality-Diversity (QD) algorithms can discover large and complex behavioural repertoires consisting of both diverse and high-performing skills. However, the generation of behavioural repertoires has mainly been limited to simulation environments instead of real-world learning. This is because existing QD algorithms need large numbers of evaluations as well as episodic resets, which require manual human supervision and interventions. This paper proposes Reset-Free Quality-Diversity optimization (RF-QD) as a step towards autonomous learning for robotics in open-ended environments. We build on Dynamics-Aware Quality-Diversity (DA-QD) and introduce a behaviour selection policy that leverages the diversity of the imagined repertoire and environmental information to intelligently select of behaviours that can act as automatic resets. We demonstrate this through a task of learning to walk within defined training zones with obstacles. Our experiments show that we can learn full repertoires of legged locomotion controllers autonomously without manual resets with high sample efficiency in spite of harsh safety constraints. Finally, using an ablation of different target objectives, we show that it is important for RF-QD to have diverse types solutions available for the behaviour selection policy over solutions optimised with a specific objective. Videos and code available at https://sites.google.com/view/rf-qd.

arXiv:2210.04964공식 코드 · Papers with Code

Generating Executable Action Plans with Environmentally-Aware Language Models

Large Language Models (LLMs) trained using massive text datasets have recently shown promise in generating action plans for robotic agents from high level text queries. However, these models typically do not consider the robot's environment, resulting in generated plans that may not actually be executable, due to ambiguities in the planned actions or environmental constraints. In this paper, we propose an approach to generate environmentally-aware action plans that agents are better able to execute. Our approach involves integrating environmental objects and object relations as additional inputs into LLM action plan generation to provide the system with an awareness of its surroundings, resulting in plans where each generated action is mapped to objects present in the scene. We also design a novel scoring function that, along with generating the action steps and associating them with objects, helps the system disambiguate among object instances and take into account their states. We evaluated our approach using the VirtualHome simulator and the ActivityPrograms knowledge base and found that action plans generated from our system had a 310% improvement in executability and a 147% improvement in correctness over prior work. The complete code and a demo of our method is publicly available at https://github.com/hri-ironlab/scene_aware_language_planner.

arXiv:2210.13826공식 코드 · Papers with Code

Instance Segmentation for Chinese Character Stroke Extraction, Datasets and Benchmarks

Stroke is the basic element of Chinese character and stroke extraction has been an important and long-standing endeavor. Existing stroke extraction methods are often handcrafted and highly depend on domain expertise due to the limited training data. Moreover, there are no standardized benchmarks to provide a fair comparison between different stroke extraction methods, which, we believe, is a major impediment to the development of Chinese character stroke understanding and related tasks. In this work, we present the first public available Chinese Character Stroke Extraction (CCSE) benchmark, with two new large-scale datasets: Kaiti CCSE (CCSE-Kai) and Handwritten CCSE (CCSE-HW). With the large-scale datasets, we hope to leverage the representation power of deep models such as CNNs to solve the stroke extraction task, which, however, remains an open question. To this end, we turn the stroke extraction problem into a stroke instance segmentation problem. Using the proposed datasets to train a stroke instance segmentation model, we surpass previous methods by a large margin. Moreover, the models trained with the proposed datasets benefit the downstream font generation and handwritten aesthetic assessment tasks. We hope these benchmark results can facilitate further research. The source code and datasets are publicly available at: https://github.com/lizhaoliu-Lec/CCSE.

arXiv:2302.10143공식 코드 · Papers with Code

Hashtag-Guided Low-Resource Tweet Classification

Social media classification tasks (e.g., tweet sentiment analysis, tweet stance detection) are challenging because social media posts are typically short, informal, and ambiguous. Thus, training on tweets is challenging and demands large-scale human-annotated labels, which are time-consuming and costly to obtain. In this paper, we find that providing hashtags to social media tweets can help alleviate this issue because hashtags can enrich short and ambiguous tweets in terms of various information, such as topic, sentiment, and stance. This motivates us to propose a novel Hashtag-guided Tweet Classification model (HashTation), which automatically generates meaningful hashtags for the input tweet to provide useful auxiliary signals for tweet classification. To generate high-quality and insightful hashtags, our hashtag generation model retrieves and encodes the post-level and entity-level information across the whole corpus. Experiments show that HashTation achieves significant improvements on seven low-resource tweet classification tasks, in which only a limited amount of training data is provided, showing that automatically enriching tweets with model-generated hashtags could significantly reduce the demand for large-scale human-labeled data. Further analysis demonstrates that HashTation is able to generate high-quality hashtags that are consistent with the tweets and their labels. The code is available at https://github.com/shizhediao/HashTation.

arXiv:2310.10480공식 코드 · Papers with Code

G-SPEED: General SParse Efficient Editing MoDel

Large Language Models~(LLMs) have demonstrated incredible capabilities in understanding, generating, and manipulating languages. Through human-model interactions, LLMs can automatically understand human-issued instructions and output the expected contents, which can significantly increase working efficiency. In various types of real-world demands, editing-oriented tasks account for a considerable proportion, which involves an interactive process that entails the continuous refinement of existing texts to meet specific criteria. Due to the need for multi-round human-model interaction and the generation of complicated editing tasks, there is an emergent need for efficient general editing models. In this paper, we propose \underline{\textbf{G}}eneral \underline{\textbf{SP}}arse \underline{\textbf{E}}fficient \underline{\textbf{E}}diting Mo\underline{\textbf{D}}el~(\textbf{G-SPEED}), which can fulfill diverse editing requirements through a single model while maintaining low computational costs. Specifically, we first propose a novel unsupervised text editing data clustering algorithm to deal with the data scarcity problem. Subsequently, we introduce a sparse editing model architecture to mitigate the inherently limited learning capabilities of small language models. The experimental outcomes indicate that G-SPEED, with its 508M parameters, can surpass LLMs equipped with 175B parameters. Our code and model checkpoints are available at \url{https://github.com/Banner-Z/G-SPEED}.

arXiv:2310.12004공식 코드 · Papers with Code

Image Super-resolution Via Latent Diffusion: A Sampling-space Mixture Of Experts And Frequency-augmented Decoder Approach

The recent use of diffusion prior, enhanced by pre-trained text-image models, has markedly elevated the performance of image super-resolution (SR). To alleviate the huge computational cost required by pixel-based diffusion SR, latent-based methods utilize a feature encoder to transform the image and then implement the SR image generation in a compact latent space. Nevertheless, there are two major issues that limit the performance of latent-based diffusion. First, the compression of latent space usually causes reconstruction distortion. Second, huge computational cost constrains the parameter scale of the diffusion model. To counteract these issues, we first propose a frequency compensation module that enhances the frequency components from latent space to pixel space. The reconstruction distortion (especially for high-frequency information) can be significantly decreased. Then, we propose to use Sample-Space Mixture of Experts (SS-MoE) to achieve more powerful latent-based SR, which steadily improves the capacity of the model without a significant increase in inference costs. These carefully crafted designs contribute to performance improvements in largely explored 4x blind super-resolution benchmarks and extend to large magnification factors, i.e., 8x image SR benchmarks. The code is available at https://github.com/amandaluof/moe_sr.

arXiv:2311.06025공식 코드 · Papers with Code

ChiMed-GPT: A Chinese Medical Large Language Model with Full Training Regime and Better Alignment to Human Preferences

Recently, the increasing demand for superior medical services has highlighted the discrepancies in the medical infrastructure. With big data, especially texts, forming the foundation of medical services, there is an exigent need for effective natural language processing (NLP) solutions tailored to the healthcare domain. Conventional approaches leveraging pre-trained models present promising results in this domain and current large language models (LLMs) offer advanced foundation for medical text processing. However, most medical LLMs are trained only with supervised fine-tuning (SFT), even though it efficiently empowers LLMs to understand and respond to medical instructions but is ineffective in learning domain knowledge and aligning with human preference. In this work, we propose ChiMed-GPT, a new benchmark LLM designed explicitly for Chinese medical domain, and undergoes a comprehensive training regime with pre-training, SFT, and RLHF. Evaluations on tasks including information extraction, question answering, and dialogue generation demonstrate ChiMed-GPT's superior performance over general domain LLMs. Furthermore, we analyze possible biases through prompting ChiMed-GPT to perform attitude scales regarding discrimination of patients, so as to contribute to further responsible development of LLMs in the medical domain. The code and model are released at https://github.com/synlp/ChiMed-GPT.

arXiv:2401.03945공식 코드 · Papers with Code

SpeechAgents: Human-Communication Simulation with Multi-Modal Multi-Agent Systems

Human communication is a complex and diverse process that not only involves multiple factors such as language, commonsense, and cultural backgrounds but also requires the participation of multimodal information, such as speech. Large Language Model (LLM)-based multi-agent systems have demonstrated promising performance in simulating human society. Can we leverage LLM-based multi-agent systems to simulate human communication? However, current LLM-based multi-agent systems mainly rely on text as the primary medium. In this paper, we propose SpeechAgents, a multi-modal LLM based multi-agent system designed for simulating human communication. SpeechAgents utilizes multi-modal LLM as the control center for individual agent and employes multi-modal signals as the medium for exchanged messages among agents. Additionally, we propose Multi-Agent Tuning to enhance the multi-agent capabilities of LLM without compromising general abilities. To strengthen and evaluate the effectiveness of human communication simulation, we build the Human-Communication Simulation Benchmark. Experimental results demonstrate that SpeechAgents can simulate human communication dialogues with consistent content, authentic rhythm, and rich emotions and demonstrate excellent scalability even with up to 25 agents, which can apply to tasks such as drama creation and audio novels generation. Code and models will be open-sourced at https://github. com/0nutation/SpeechAgents