Projects

### Overview: 每日更新 LLM、Agent 相关研究论文，提供 arXiv 地址、GitHub 仓库和文章总结。 ### Key features: - EpiSEM-Bench: 评估 LLM 智能体在证据缺失情况下的认识论安全。 - AgenticEvolve: 通过演化搜索循环，解决模型缺乏代码执行和分析能力的问题。 - Beyond Accuracy: 分解任务与类型化契约在多智能体 LLM 流水线中的作用。 - AfterDeleteBench: 评估 LLM 智能体在删除原始记忆记录后信息恢复能力。 ### Tech stack: Python ### Why it matters: 提供最新的 LLM 和 Agent 相关研究论文，帮助开发者和研究人员了解最新进展和挑战。

### Overview: MagicTime is a novel metamorphic video generation pipeline designed to enhance the realism of generated time-lapse videos. It leverages the power of prompts to create highly accurate and contextually relevant videos, making it a significant advancement in the field of video synthesis. ### Key Features: - **Metamorphic Simulators**: MagicTime uses prompts to simulate real-world scenarios, ensuring that the generated videos are as realistic as possible. - **Prompt-Driven Generation**: The pipeline is designed to generate videos that closely match the input prompts, making it highly effective for tasks requiring high accuracy. - **Realistic Scenarios**: It supports a wide range of scenarios, including but not limited to, time-lapse videos, animations, and other creative applications. ### Tech Stack: MagicTime is implemented using Python, leveraging the capabilities of the Hugging Face Transformers library for model training and inference. ### Why It Matters: MagicTime represents a significant leap forward in the field of video generation, particularly in its ability to simulate real-world scenarios with high accuracy. This technology has the potential to revolutionize various applications, including time-lapse video creation, animation, and more, by providing a more realistic and contextually relevant output. ### Repository: The MagicTime repository is available on GitHub, where you can find the latest updates and contributions. It is also compatible with the Hugging Face Spaces, allowing for seamless integration with other tools and platforms. ### GitHub Stars: The repository has received 1,336 stars from the developer community, indicating a high level of interest and engagement in the project. ###

### Overview: Lean Copilot is a project that leverages large language models (LLMs) to automate theorem proving in the Lean theorem prover. It allows users to suggest tactics, provide proof suggestions, and search for proofs directly within Lean, enhancing the efficiency and effectiveness of theorem proving. ### Key Features: - **Tactic Suggestion**: Provides suggestions for tactics and premises. - **Proof Search**: Facilitates the search for proofs. - **Premise Selection**: Helps in selecting appropriate premises for proofs. ### Tech Stack: - **Lean**: The programming language used for theorem proving. - **LLMs (Large Language Models)**: Used to automate the proof process. - **LeanDojo**: A community-driven project for Lean development and hosting. ### Why It Matters: Lean Copilot streamlines the theorem proving process by integrating AI-driven suggestions and automation, making the proofing experience more efficient and accessible. It leverages the power of AI to enhance the capabilities of the Lean theorem prover, making it more powerful and user-friendly. ### Repository: The repository for Lean Copilot is hosted on GitHub at [lean-dojo/LeanCopilot](https://github.com/lean-dojo/LeanCopilot). It has received 1,334 GitHub stars, indicating a significant interest and adoption in the project. ### Table of Contents: 1. **Requirements** 2. **Using Lean Copilot in Your Project** - **Adding Lean Copilot as a Dependency** - **Getting Started with Lean Copilot** - **Tactic Suggestion** -

Overview: This repository contains the implementation of Natural Speech 2, a zero-shot speech and singing synthesizer system, utilizing Pytorch. It leverages a neural audio codec and a latent diffusion model for natural and zero-shot text-to-speech synthesis. Key features: - Zero-shot Speech and Singing Synthesizer - Neural Audio Codec with Continuous Latent Vectors - Non-Autoregressive Generation with Latent Diffusion Model Tech stack: Pytorch Why it matters: This project showcases advancements in artificial intelligence for speech synthesis, offering a practical tool for developers and researchers interested in natural language processing and audio generation. Repository: lucidrains/naturalspeech2-pytorch GitHub stars: 1334

Overview: The TensorFlow Object Counting API is an open-source framework designed to simplify the development of object counting systems using TensorFlow and Keras. Key features: - Cumulative Counting Mode - Real-Time Counting Mode - Object Tracking Mode - Object Counting on Single Image - Object Counting based R-CNN - Object Segmentation & Counting based Mask R-CNN - Custom Object Counting Mode Tech stack: Python Why it matters: This framework provides a user-friendly interface for developing object counting systems, making it accessible to developers without extensive machine learning expertise.

### Overview This repository contains a collection of papers from the Natural Language Processing (NLP) domain, including reading notes and code implementations for various models and techniques. The repository also includes tools for text similarity and pre-trained models, as well as a search tool for easy access to the papers. ### Key Features - **Reading Notes and Code**: Includes reading notes and code implementations for various NLP models and techniques. - **Paper Replication and Implementation**: Provides code for reproducing models and implementing various NLP techniques. - **Search Tool**: A tool for searching through the papers and their associated code. ### Tech Stack - **Python** - The primary programming language used for the implementation and code. ### Why It Matters This repository serves as a comprehensive resource for anyone interested in NLP, providing both theoretical insights and practical implementations. It is particularly useful for researchers and developers looking to understand and implement various NLP techniques and models. ### GitHub Stars - **Stars**: 1332 ### Repository - **Repository**: DengBoCong/nlp-paper - **Primary Language**: Python - **GitHub Stars**: 1332 ### Repository Links - **Paper Repository**: [NLP-Paper](https://github.com/DengBoCong/nlp-paper) - **Search Tool**: [Search Kits](https://github.com/DengBoCong/nlp-paper/blob/master/paper-code/image/preview.gif) - **Code Repository**: [paper-code](https://github.com/DengBoCong/paper/tree/master/paper-code) ### Contents

### Overview: Weave is an entity-level git merge driver that resolves false conflicts by merging code into entities (functions, classes, and keys) instead of lines. This approach significantly reduces false conflicts and improves developer productivity. ### Key Features: - **Entity-Level Merge**: Weave parses code into entities and merges those entities instead of lines. - **95% Reduction in Conflicts**: Weave reduces false conflicts by 95% compared to line-based merge. - **Agent-Native Infrastructure**: Weave is part of the Ataraxy Labs stack and integrates with semantic version control, code review, and agent sessions. ### Tech Stack: Rust ### Why It Matters: Weave resolves false conflicts that Git cannot, improving developer productivity and reducing human intervention in the development process. This makes it particularly useful in environments where multiple AI agents are working on the same codebase. ### Repository: Repository: Ataraxy-Labs/weave GitHub Stars: 1331 ### Primary Language: Rust ### GitHub Description: "Entity-level semantic merge for Git. Resolves false conflicts Git invents when independent agents edit the same file. ~95% reduction vs. line-based merge." ### Tech Stack: Rust ### Why It Matters: Weave resolves false conflicts that Git cannot, improving developer productivity and reducing human intervention in the development process. This makes it particularly useful in environments where multiple AI agents are working on the same codebase.

Overview: OpenPets is a local first, desktop companion platform featuring animated pets that can be enhanced with plugins and optional local agent integrations. It offers a variety of features including focus timers, reminders, mood check-ins, mini games, and more. Key features: - Animated desktop pets - Official plugins for various functionalities - Plugin SDK v3 for custom plugin development - Optional local agent layer for additional interactions Tech stack: TypeScript Why it matters: OpenPets transforms your desktop into a dynamic companion, making productivity and personalization more engaging and interactive. Repository: OpenPetsHQ/openpets GitHub stars: 1320 GitHub description: Local first, desktop companion platform with animated pets, plugins, and optional local agent integrations.

### Overview: Meta Ads MCP is a hosted remote Model Context Protocol (MCP) server that enables the seamless integration of AI assistants like Claude, ChatGPT, Perplexity, and Cursor into Meta Ads campaigns across Facebook, Instagram, and every Meta ad surface. This MCP node is part of the Pipeboard MCP family, offering a unified interface with a single login for Meta, Google, TikTok, Snap, and Reddit ads, all managed through a single auth and safety model. ### Key Features: - **Unified Interface**: A single login for all Meta, Google, TikTok, Snap, and Reddit ads. - **Seamless Integration**: Runs AI assistants Claude, ChatGPT, Perplexity, and Cursor across Meta Ads. - **Hosted Remote**: No self-hosting required, hosted on Pipeboard. - **230+ Tools**: Includes 42 tools for Meta Ads, 59 for Google Ads, 59 for TikTok Ads, 37 for Snap Ads, and 33 for Reddit Ads. - **Campaign Management**: Launch, update, and analyze campaigns. - **Ad Set and Ad Management**: Create, update, and track ad sets and ads. - **Creative Management**: Upload and manage creatives, including dynamic creative testing. - **Insights and Analytics**: Access insights and performance reports. - **Surface-Specific Tools**: Tailored tools for each ad platform, such as Facebook, Instagram, TikTok, Snap, and Reddit. ### Tech Stack: Python ### Why It Matters: Meta Ads

Overview: Bernstein is an open-source governance layer designed for AI agents, enabling declarative rule writing and verifiable, replayable record enforcement. Key features: - Declarative rule writing - Verifiable, replayable record enforcement - Free and Apache-2.0 licensed Tech stack: Python Why it matters: Bernstein simplifies the process of creating and enforcing rules for AI agents, making it easier and more transparent for developers to manage governance and decision-making processes.

Overview: AlexaPi is a versatile client for Amazon's Alexa service, designed to run on various platforms including Raspberry Pi, Orange Pi, and Linux/Windows desktops. It aims to provide a seamless experience for voice commands. Key features: - Supports multiple platforms including Raspberry Pi, Orange Pi, and Linux/Windows systems. - Compatible with a wide range of audio peripherals for better integration. - Optional push button and dual color LED for enhanced user interaction. Tech stack: Python Why it matters: AlexaPi offers a convenient and efficient way to control smart home devices using voice commands, making it a valuable tool for automation enthusiasts and tech enthusiasts alike. Repository: alexa-pi/AlexaPi Primary language: Python GitHub stars: 1327 GitHub description: Alexa client for all your devices! # No active development. PRs welcome # consider https://github.com/respeaker/avs instead

Overview: Hermes Atlas is a comprehensive directory of tools, skills, and integrations for the Hermes Agent, offering a curated catalog of 240+ ecosystem resources with live GitHub data. Key features: - 80+ quality-filtered repos across 12 categories - Live star counts fetched from GitHub API and cached in Redis - Sparklines and trending badges showing growth over time - RAG-powered chatbot answering questions about Hermes Agent - Search, sort, filter across the entire ecosystem - Light and dark mode with OS preference detection - Mobile responsive Tech stack: Vanilla HTML/CSS/JavaScript, Vercel static + serverless functions, Redis Cloud, OpenAI text-embedding-3-small, GPT-6 Luna, Claude Haiku 5.5 Why it matters: Hermes Atlas serves as the primary resource for developers and enthusiasts of the Hermes Agent ecosystem, providing a curated and up-to-date directory of tools and integrations.

### Overview: AntSK 是一个基于 .NET 9、Blazor、Semantic Kernel 和 Kernel Memory 的企业级 AI 知识库与智能体平台，旨在为企业提供强大的 AI 知识管理解决方案。它支持本地离线运行，提供国产信创适配，适用于金融、政务、医疗、军工等敏感数据场景。 ### Key Features: 1. **本地离线运行**：支持 Ollama、LLamaFactory 和 llama.cpp(GGUF) 三类本地推理引擎，实现完全断网运行。 2. **数据不出内网**：满足金融、政务、医疗、军工等数据敏感场景。 3. **软硬一体 AI 一体机**：提供 32B / 70B 梯度配置的 AI 一体机，适配主流开源模型，开箱即用。 4. **国产信创适配**：兼容国产大模型与国产数据库（含达梦）。 5. **零迁移成本二次开发**：C# 团队可零迁移成本二次开发，并享受强类型与编译期检查带来的稳定性。 ### Tech Stack: .NET 9、Blazor Server、ASP.NET Core、Semantic Kernel、Kernel Memory、SqlSugar ORM、Ant Design Blazor ### Why it Matters: - **填补 .NET 生态空白**：提供成熟的 .NET / C# 企业级 AI 知识库解决方案，适合 C# 团队快速开发与二次开发。 - **数据不出内网**：满足金融

Overview: This project is a comprehensive survey of large language model (LLM)-based multi-agents, focusing on their architectures, applications, and challenges. The survey is based on papers presented at the 2024 International Joint Conference on Artificial Intelligence (IJCAI). Key features: - **Survey of LLM-based Multi-Agents**: Comprehensive overview of architectures and applications of multi-agent systems using LLMs. - **Architecture Diagrams**: Visual representations of LLM-based multi-agent systems. - **Citation Information**: Provides a table of contents with links to the survey paper and code repository. Tech stack: The project uses Markdown, GitHub, and various LaTeX packages for document formatting. Why it matters: This survey is crucial for researchers and practitioners interested in the field of multi-agent systems and language models, providing a clear understanding of the current state of the art and future challenges.

Overview: Bert4torch is an elegant PyTorch implementation of transformers, designed to simplify the process of loading and fine-tuning various pre-trained models for natural language processing tasks. Key features: - Load and fine-tune pre-trained models like ChatGLM, LLaMA, Baichuan, Ziya, Bloom, etc., using a single command. - Load and fine-tune pre-trained models like BERT, ROBERTA, ALBERT, XLNet, NEZHA, Bart, RoFormer, RoFormer_V2, ELECTRA, GPT, GPT2, T5, GAU-alpha, ERNIE, etc., for fine-tuning and further improvements. - Provide a variety of examples including LLM, pretraining, sentence classification, sentence embedding, sequence labeling, relation extraction, seq2seq, and serving. - Includes common training tricks for easy integration. - Supports loading transformers models together. - Offers a training progress bar, concise call syntax, and efficient parameter information printing. - Provides a default logger and Tensorboard for training process recording. - Allows for customizing the training process to meet high requirements. Tech stack: Python Why it matters: Bert4torch simplifies the process of loading and fine-tuning pre-trained models for NLP tasks, making it easier for developers to start working on their projects without needing to dive into the complexities of PyTorch or transformers directly. Repository: Tongjilibo/bert4torch Primary language: Python GitHub stars: 1326 GitHub description: An elegent py

Overview: Your Smart Second Brain is a free, open-source Obsidian plugin that enhances your vault with better search, an interactive knowledge graph, and an AI assistant that understands your notes. Key features: - **Search**: Combines keyword matching with meaning, enabling you to find the note you were thinking of rather than the one you typed. - **Smart Graph**: Groups your vault's notes into topics based on links, tags, and what you wrote, providing semantic search and insights. - **Agents**: Reads and writes your notes, drawing charts and dashboards live from them, and integrates with other plugins. Tech stack: TypeScript Why it matters: This plugin transforms Obsidian into a smarter, more interactive knowledge management tool, enhancing your workflow and productivity. Repository: s2b-dev/smart-second-brain GitHub stars: 1326 GitHub description: A free, open-source Obsidian plugin that makes your vault smarter: better search, an interactive knowledge graph, and an AI assistant that knows your notes.

### Overview: Pruna is a model optimization framework designed for developers, enabling them to deliver faster, more efficient models with minimal overhead. ### Key features: - **Faster**: Accelerate inference times through advanced optimization techniques. - **Smaller**: Reduce model size while maintaining quality. - **Cheaper**: Lower computational costs and resource requirements. - **Greener**: Decrease energy consumption and environmental impact. ### Tech stack: Python ### Why it matters: Pruna is a comprehensive suite of compression algorithms, including caching, quantization, pruning, distillation, and compilation techniques, designed to optimize model performance without compromising quality. It simplifies the process of optimizing models, requiring just a few lines of code, making it accessible to developers of all skill levels.

### Overview: Bedrock Chat is a multilingual generative AI platform powered by Amazon Bedrock, designed for developers to create chatbots and knowledge-based bots. It supports chat functionality, custom bot creation with knowledge retrieval (RAG), bot sharing through a bot store, and task automation using agents. ### Key Features: - **Chatbot Creation**: Easily set up chatbots that can handle customer inquiries and provide personalized responses. - **Knowledge-Based Bots**: Implement bots that leverage Amazon Bedrock's knowledge retrieval capabilities for enhanced knowledge-based interactions. - **Bot Sharing**: Share bots across multiple applications via a bot store, allowing for easy integration and management. - **Task Automation**: Automate tasks using agents, streamlining workflows and processes. ### Tech Stack: TypeScript ### Why It Matters: Bedrock Chat simplifies the creation and management of AI-driven chatbots and knowledge-based bots, making it easier for developers to integrate advanced AI capabilities into their applications. The platform's support for multiple languages and its ability to share bots across applications make it a valuable tool for businesses looking to enhance their customer service and knowledge management capabilities.

### Overview: This project provides an overview of recent trends in deep learning based natural language processing (NLP), covering theoretical descriptions and implementation details of deep learning models such as recurrent neural networks (RNNs), convolutional neural networks (CNNs), and reinforcement learning. It includes state of the art results, emerging concepts, new benchmark datasets, and code/dataset releases, aiming to maintain an up-to-date learning resource for researchers and enthusiasts interested in modern techniques applied to NLP. ### Key features: - State of the art results - Emerging concepts and applications - New benchmark datasets - Code/dataset releases - Friendly and open resource for NLP research ### Tech stack: CSS ### Why it matters: This project serves as a comprehensive guide for anyone interested in modern techniques applied to NLP, offering valuable insights into the latest trends and methodologies used in the field.

### Overview: Kusion is a declarative intent-driven Platform Orchestrator designed to streamline the development process for internal developers. It enables app-centric development by allowing developers to specify the workload and resource dependencies in a single configuration, eliminating the need for environment-specific configurations. ### Key Features: - **AppConfiguration**: A single configuration that defines the workload and resource dependencies. - **Intent-Driven Orchestration**: Ensures that the necessary resources are provisioned automatically based on the specified configuration. - **Zero-Configuration Setup**: Developers only need to write a single configuration, reducing the complexity and effort required for setup. ### Tech Stack: Go ### Why It Matters: Kusion simplifies the development process by abstracting away the complexities of environment-specific configurations, allowing developers to focus on writing applications. This results in faster development cycles and reduced operational overhead, making it an essential tool for modern internal developer platforms.