EvoCodeBench: An Evolving Code Generation Benchmark Aligned with Real-World Code Repositories

How to evaluate Large Language Models (LLMs) in code generation is an open question. Existing benchmarks demonstrate poor alignment with real-world code repositories and are insufficient to evaluate the coding abilities of LLMs. This paper proposes a new benchmark - EvoCodeBench to address the preceding problems, which has three primary advances. (1) EvoCodeBench aligns with real-world repositories in multiple dimensions, e.g., code distributions and dependency distributions. (2) EvoCodeBench offers comprehensive annotations (e.g., requirements, reference code, and reference dependencies), and robust evaluation metrics (e.g., Pass@k and Recall@k). (3) EvoCodeBench is an evolving benchmark to avoid data leakage. We build an automatic pipeline to update EvoCodeBench from the latest repositories. We release the first version - EvoCodeBench-2403, containing 275 samples from 25 real-world repositories. Based on EvoCodeBench, we propose repository-level code generation and evaluate 10 popular LLMs (e.g., gpt-4, gpt-3.5, DeepSeek Coder, StarCoder 2, CodeLLaMa, Gemma, and Qwen 1.5). Our experiments reveal the coding abilities of these LLMs in real-world repositories. For example, the highest Pass@1 of gpt-4 only is 20.73% in our experiments. We also analyze failed cases and summarize the shortcomings of existing LLMs in EvoCodeBench. We release EvoCodeBench, all prompts, and LLMs' completions for further community analysis.

DiffMAC: Diffusion Manifold Hallucination Correction for High Generalization Blind Face Restoration

Blind face restoration (BFR) is a highly challenging problem due to the uncertainty of degradation patterns. Current methods have low generalization across photorealistic and heterogeneous domains. In this paper, we propose a Diffusion-Information-Diffusion (DID) framework to tackle diffusion manifold hallucination correction (DiffMAC), which achieves high-generalization face restoration in diverse degraded scenes and heterogeneous domains. Specifically, the first diffusion stage aligns the restored face with spatial feature embedding of the low-quality face based on AdaIN, which synthesizes degradation-removal results but with uncontrollable artifacts for some hard cases. Based on Stage I, Stage II considers information compression using manifold information bottleneck (MIB) and finetunes the first diffusion model to improve facial fidelity. DiffMAC effectively fights against blind degradation patterns and synthesizes high-quality faces with attribute and identity consistencies. Experimental results demonstrate the superiority of DiffMAC over state-of-the-art methods, with a high degree of generalization in real-world and heterogeneous settings. The source code and models will be public.

EventRPG: Event Data Augmentation with Relevance Propagation Guidance

Event camera, a novel bio-inspired vision sensor, has drawn a lot of attention for its low latency, low power consumption, and high dynamic range. Currently, overfitting remains a critical problem in event-based classification tasks for Spiking Neural Network (SNN) due to its relatively weak spatial representation capability. Data augmentation is a simple but efficient method to alleviate overfitting and improve the generalization ability of neural networks, and saliency-based augmentation methods are proven to be effective in the image processing field. However, there is no approach available for extracting saliency maps from SNNs. Therefore, for the first time, we present Spiking Layer-Time-wise Relevance Propagation rule (SLTRP) and Spiking Layer-wise Relevance Propagation rule (SLRP) in order for SNN to generate stable and accurate CAMs and saliency maps. Based on this, we propose EventRPG, which leverages relevance propagation on the spiking neural network for more efficient augmentation. Our proposed method has been evaluated on several SNN structures, achieving state-of-the-art performance in object recognition tasks including N-Caltech101, CIFAR10-DVS, with accuracies of 85.62% and 85.55%, as well as action recognition task SL-Animals with an accuracy of 91.59%. Our code is available at https://github.com/myuansun/EventRPG.

All in One: Multi-Task Prompting for Graph Neural Networks (Extended Abstract)

This paper is an extended abstract of our original work published in KDD23, where we won the best research paper award (Xiangguo Sun, Hong Cheng, Jia Li, Bo Liu, and Jihong Guan. All in one: Multi-task prompting for graph neural networks. KDD 23) The paper introduces a novel approach to bridging the gap between pre-trained graph models and the diverse tasks they're applied to, inspired by the success of prompt learning in NLP. Recognizing the challenge of aligning pre-trained models with varied graph tasks (node level, edge level, and graph level), which can lead to negative transfer and poor performance, we propose a multi-task prompting method for graphs. This method involves unifying graph and language prompt formats, enabling NLP's prompting strategies to be adapted for graph tasks. By analyzing the task space of graph applications, we reformulate problems to fit graph-level tasks and apply meta-learning to improve prompt initialization for multiple tasks. Experiments show our method's effectiveness in enhancing model performance across different graph tasks. Beyond the original work, in this extended abstract, we further discuss the graph prompt from a bigger picture and provide some of the latest work toward this area.

GraphWiz: An Instruction-Following Language Model for Graph Problems

Large language models (LLMs) have achieved impressive success across several fields, but their proficiency in understanding and resolving complex graph problems is less explored. To bridge this gap, we introduce GraphInstruct, a novel and comprehensive instruction-tuning dataset designed to equip language models with the ability to tackle a broad spectrum of graph problems using explicit reasoning paths. Utilizing GraphInstruct, we build GraphWiz, an open-source language model capable of resolving various graph problem types while generating clear reasoning processes. To enhance the model's capability and reliability, we incorporate the Direct Preference Optimization (DPO) framework into the graph problem-solving context. The enhanced model, GraphWiz-DPO, achieves an average accuracy of 65% across nine tasks with different complexity levels, surpassing GPT-4 which has an average accuracy of 43.8%. Moreover, our research delves into the delicate balance between training data volume and model performance, highlighting the potential for overfitting with increased data. We also explore the transferability of the model's reasoning ability across different graph tasks, indicating the model's adaptability and practical application potential. Our investigation offers a new blueprint and valuable insights for developing LLMs specialized in graph reasoning and problem-solving.

StarCoder 2 and The Stack v2: The Next Generation

The BigCode project, an open-scientific collaboration focused on the responsible development of Large Language Models for Code (Code LLMs), introduces StarCoder2. In partnership with Software Heritage (SWH), we build The Stack v2 on top of the digital commons of their source code archive. Alongside the SWH repositories spanning 619 programming languages, we carefully select other high-quality data sources, such as GitHub pull requests, Kaggle notebooks, and code documentation. This results in a training set that is 4x larger than the first StarCoder dataset. We train StarCoder2 models with 3B, 7B, and 15B parameters on 3.3 to 4.3 trillion tokens and thoroughly evaluate them on a comprehensive set of Code LLM benchmarks. We find that our small model, StarCoder2-3B, outperforms other Code LLMs of similar size on most benchmarks, and also outperforms StarCoderBase-15B. Our large model, StarCoder2- 15B, significantly outperforms other models of comparable size. In addition, it matches or outperforms CodeLlama-34B, a model more than twice its size. Although DeepSeekCoder- 33B is the best-performing model at code completion for high-resource languages, we find that StarCoder2-15B outperforms it on math and code reasoning benchmarks, as well as several low-resource languages. We make the model weights available under an OpenRAIL license and ensure full transparency regarding the training data by releasing the SoftWare Heritage persistent IDentifiers (SWHIDs) of the source code data.

Compress to Impress: Unleashing the Potential of Compressive Memory in Real-World Long-Term Conversations

Existing retrieval-based methods have made significant strides in maintaining long-term conversations. However, these approaches face challenges in memory database management and accurate memory retrieval, hindering their efficacy in dynamic, real-world interactions. This study introduces a novel framework, COmpressive Memory-Enhanced Dialogue sYstems (COMEDY), which eschews traditional retrieval modules and memory databases. Instead, COMEDY adopts a ''One-for-All'' approach, utilizing a single language model to manage memory generation, compression, and response generation. Central to this framework is the concept of compressive memory, which intergrates session-specific summaries, user-bot dynamics, and past events into a concise memory format. To support COMEDY, we curated a large-scale Chinese instruction-tuning dataset, Dolphin, derived from real user-chatbot interactions. Comparative evaluations demonstrate COMEDY's superiority over traditional retrieval-based methods in producing more nuanced and human-like conversational experiences. Our codes are available at https://github.com/nuochenpku/COMEDY.

ZeroG: Investigating Cross-dataset Zero-shot Transferability in Graphs

With the development of foundation models such as large language models, zero-shot transfer learning has become increasingly significant. This is highlighted by the generative capabilities of NLP models like GPT-4, and the retrieval-based approaches of CV models like CLIP, both of which effectively bridge the gap between seen and unseen data. In the realm of graph learning, the continuous emergence of new graphs and the challenges of human labeling also amplify the necessity for zero-shot transfer learning, driving the exploration of approaches that can generalize across diverse graph data without necessitating dataset-specific and label-specific fine-tuning. In this study, we extend such paradigms to zero-shot transferability in graphs by introducing ZeroG, a new framework tailored to enable cross-dataset generalization. Addressing the inherent challenges such as feature misalignment, mismatched label spaces, and negative transfer, we leverage a language model to encode both node attributes and class semantics, ensuring consistent feature dimensions across datasets. We also propose a prompt-based subgraph sampling module that enriches the semantic information and structure information of extracted subgraphs using prompting nodes and neighborhood aggregation, respectively. We further adopt a lightweight fine-tuning strategy that reduces the risk of overfitting and maintains the zero-shot learning efficacy of the language model. The results underscore the effectiveness of our model in achieving significant cross-dataset zero-shot transferability, opening pathways for the development of graph foundation models. Especially, ZeroG, as a zero-shot method, can even achieve results comparable to those of semi-supervised learning on Pubmed.

All in One and One for All: A Simple yet Effective Method towards Cross-domain Graph Pretraining

Large Language Models (LLMs) have revolutionized the fields of computer vision (CV) and natural language processing (NLP). One of the most notable advancements of LLMs is that a single model is trained on vast and diverse datasets spanning multiple domains -- a paradigm we term `All in One'. This methodology empowers LLMs with super generalization capabilities, facilitating an encompassing comprehension of varied data distributions. Leveraging these capabilities, a single LLM demonstrates remarkable versatility across a variety of domains -- a paradigm we term `One for All'. However, applying this idea to the graph field remains a formidable challenge, with cross-domain pretraining often resulting in negative transfer. This issue is particularly important in few-shot learning scenarios, where the paucity of training data necessitates the incorporation of external knowledge sources. In response to this challenge, we propose a novel approach called Graph COordinators for PrEtraining (GCOPE), that harnesses the underlying commonalities across diverse graph datasets to enhance few-shot learning. Our novel methodology involves a unification framework that amalgamates disparate graph datasets during the pretraining phase to distill and transfer meaningful knowledge to target tasks. Extensive experiments across multiple graph datasets demonstrate the superior efficacy of our approach. By successfully leveraging the synergistic potential of multiple graph datasets for pretraining, our work stands as a pioneering contribution to the realm of graph foundational model.