Projects

### Overview: This repository contains a survey paper on the cutting-edge research on reasoning with language model prompting. It aims to provide a comprehensive overview of recent advancements, comparisons, and systematic resources for beginners. The paper discusses various methods, including strategy enhanced reasoning, knowledge enhanced reasoning, and external engines such as physical simulators and code interpreters. The survey also explores the potential reasons for the emergence of such reasoning abilities and highlights future research directions. ### Key features: - **Overview**: A comprehensive survey of cutting-edge research on reasoning with language model prompting. - **Methods**: - **Strategy Enhanced Reasoning**: - **Prompt Engineering**: - **Single-Stage**: - **Multi-Stage**: - **Process Optimization**: - **Self-Optimization**: - **Ensemble-Optimization**: - **Iterative-Optimization**: - **Knowledge Enhanced Reasoning**: - **Implicit Knowledge**: - **Explicit Knowledge**: - **Others**: - **External Engines**: - **Physical Simulator**: - **Code Interpreter**: - **Tool Learning**: - **Analysis**: - **Resources**: - **Benchmarks and Tasks**: - **Tools**: ### Tech stack: Unknown ### Why it matters: This survey paper provides a valuable resource for researchers and practitioners looking to understand and leverage the capabilities of language models for reasoning tasks. It serves as a foundational guide for beginners and highlights the potential for further

### Overview: These skills are designed to teach your AI coding agent how to create motion graphics, animation, and video. The skills are open-source and can be used to enhance the motion capabilities of your AI agent, making it more versatile and capable of handling complex motion tasks. ### Key features: - **TikTok/Reels/Shorts Creators**: 4 skills for TikTok/Reels/Shorts creators. - **Text-Message Video Creators**: 1 skill for text-message story creators. - **Podcasters & YouTubers**: 2 skills for podcasters and YouTubers. - **E-commerce Sellers**: 3 skills for e-commerce sellers. - **Brand Advertisers & Marketers**: 3 skills for brand advertisers and marketers. - **Analysts & PMs**: 3 skills for analysts and project managers. - **Educators & Content Creators**: 5 skills for educators and content creators. - **Vox-style Editorial Maps**: 1 skill for creating Vox-style editorial maps. - **Frontend Developers**: 9 skills for frontend developers. - **Motion Designers**: 9 skills for motion designers. - **Kinetic Typography**: 1 skill for kinetic type/text animation. - **Freelancers & Studios**: 5 skills for freelancers and studios. - **WebGL Animation**: 3 skills for 3D/WebGL/technical. - **Math / Educational Animators**: 1 skill for math and educational animators. - **Code-Drawn Canvas Animation**: 2 skills for

Overview: This 10-week, 30-minute-a-day roadmap provides engineers with a practical approach to running Large Language Models (LLMs) in production, not just theoretical knowledge. Key features: - Deploying an OpenAI-compatible inference service - Instrumenting and optimizing the service - Quantizing and speculative decoding - Benchmarking across FP16, FP8, INT4, and speculative decoding - Implementing a cost-aware router Tech stack: Python, transformers, command line, Kubernetes, CUDA Why it matters: This approach ensures that you have a fully operational and reproducible solution, rather than just a collection of disconnected experiments. Repository: patchy631/time-to-first-token Primary language: HTML GitHub stars: 995 GitHub description: A 10-week, 30-minutes-a-day roadmap for LLM inference serving and optimization. vLLM, SGLang, quantization, speculative decoding, benchmarking.

### Overview: Efficient-Segmentation-Networks is a comprehensive PyTorch implementation of lightweight semantic segmentation models optimized for real-time performance. This project includes a variety of state-of-the-art models such as SQNet, LinkNet, SegNet, UNet, ENet, ERFNet, EDANet, ESPNet, ESPNetv2, LEDNet, FSSNet, CGNet, DABNet, Fast-SCNN, ContextNet, FPENet, and many others. The models are designed to be efficient, fast, and suitable for real-time applications. ### Key Features: - **Models**: Supports multiple advanced semantic segmentation models, including SQNet, LinkNet, SegNet, UNet, ENet, ERFNet, EDANet, ESPNet, ESPNetv2, LEDNet, FSSNet, CGNet, DABNet, Fast-SCNN, ContextNet, FPENet, and more. - **Efficiency**: Optimized for real-time performance, making it suitable for applications requiring quick processing times. - **Lightweight**: Designed to be lightweight, reducing computational and memory requirements. - **Modifiability**: The models are highly customizable, allowing users to fine-tune parameters for specific use cases. - **Loss Functions**: Supports various loss functions, including Weighted Cross Entropy, Weighted Cross Entropy with Label Smooth, Focal Loss, Ohem Cross Entropy, LovaszSoftmax, SegLoss-List, and more. - **Optimizers**: Offers support for popular optimizers,

### Overview: Each use case in this curated collection includes a clear introduction explaining what it does and why it matters. It also provides links to skills on ClawdHub for those needing assistance, detailed setup prompts, and measurable outcomes to track progress. ### Key Features: - **Introduction**: Detailed descriptions of each use case. - **Skills Needed**: Links to relevant skills on ClawdHub. - **Setup Instructions**: Clear prompts and configuration steps. - **Success Metrics**: Measurable outcomes to evaluate performance. ### Tech Stack: The use cases are based on real-world implementations on Moltbook, leveraging OpenClaw AI agents. ### Why It Matters: These use cases demonstrate practical applications of OpenClaw AI agents in everyday life, making them accessible to anyone without requiring coding or technical expertise. The curated nature ensures each use case is unique and valuable, providing a comprehensive collection of real-world use cases for OpenClaw AI agents.

### Overview: Each use case in this curated collection of 74 concrete, non-repeating OpenClaw AI agent use cases is designed to be easily set up and run without requiring advanced coding or technical knowledge. These scenarios are sourced from real-world implementations on Moltbook and offer practical applications for everyday life, content creation, cost optimization, and more. ### Key Features: - **Introduction**: Detailed descriptions of each use case, explaining what it does and why it matters. - **Skills You Need**: Links to ClawdHub skills that are required to set up and run the use cases. - **How to Setup**: Step-by-step prompts and configuration instructions to quickly deploy the use cases. - **Success Metrics**: Measurable outcomes that can be tracked and evaluated to determine the effectiveness of the use cases. ### Tech Stack: The use cases are designed to work with OpenClaw AI agents, leveraging the capabilities of the platform to automate tasks and improve efficiency. ### Why It Matters: These use cases are crucial for anyone looking to streamline their workflow, automate repetitive tasks, or enhance their productivity. Whether you're a business professional, a content creator, or simply someone looking to improve their daily routine, these use cases provide practical solutions that can help you achieve your goals more efficiently. ### Repository: The repository for this collection is located at [GitHub](https://github.com/EvoLinkAI/awesome-openclaw-usecases-moltbook). It has been starred by 1,006 users, indicating its popularity and usefulness in the community. ### Primary Language:

Overview: 天津大学课程共享计划是一个旨在为学生提供公开、易于获取和共同完善的课程资料和经验的项目。它通过GitHub项目作为平台，收录了选课攻略、电子教材、课件、平时作业、历年试卷、复习资料等各类课程资源，帮助学生在大学学习过程中节省时间和精力，提高学习效率。 Key features: - 收录多种课程资源，包括选课攻略、电子教材、课件、平时作业、历年试卷、复习资料等。 - 项目收录了天津大学多个课程的详细说明和经验分享，旨在帮助学生在大学学习过程中节省时间和精力，提高学习效率。 - GitHub项目作为平台，方便用户进行文件组织和协作，同时提供Issue和Pull Request等功能，便于贡献者和维护者进行交流和维护。 Tech stack: 使用Tcl语言编写，GitHub作为项目存储和协作平台。 Why it matters: 本项目旨在解决学生在大学学习过程中遇到的资料收集和共享问题，通过公开和易于获取的资源，帮助学生节省时间和精力，提高学习效率，同时促进知识的共享和积累。

### Overview: Summarization Papers is a repository containing academic papers on summarization techniques, focusing on abstractive and generative methods. The repository includes papers from various conferences and workshops, covering topics such as dialogue summarization, cross-lingual summarization, and more. ### Key Features: - **Papers**: A collection of over 100 papers on summarization techniques. - **Authors**: Contributions from multiple researchers including Yichong Huang, Haozheng Yang, and Jiaan Wang. - **Summarization Learning Route**: A detailed route map for summarization learning, providing insights into the theoretical foundations and practical applications. ### Tech Stack: - **TeX**: The primary programming language used for documentation and presentations. ### Why It Matters: Summarization Papers serves as a valuable resource for researchers and practitioners in the field of natural language processing, offering comprehensive insights into cutting-edge summarization techniques and their applications across different domains. ### GitHub Stars: - **GitHub Stars**: 1005, indicating significant interest and engagement with the repository. ### Contributors: - **Contributors**: Yichong Huang, Haozheng Yang, and Jiaan Wang. ### Summarization Learning Route: - A detailed route map for summarization learning, providing theoretical foundations and practical applications. ### Trending: - **Trending**: Displays the most recent trends and updates in summarization research. ### Presentations and Notes: - **Presentations**: Links to presentations and slides from various conferences and workshops. - **Notes**: Links to blog posts and notes on summarization techniques

Overview: DoRA is an official PyTorch implementation of Weight-Decomposed Low-Rank Adaptation (DoRA), a method that decomposes pre-trained weights into magnitude and direction components for efficient fine-tuning on large language models. It enhances the learning capacity and training stability of LoRA while reducing the number of trainable parameters. Key features: - Decomposes pre-trained weights into magnitude and direction components. - Employing LoRA for directional updates to minimize the number of trainable parameters efficiently. - Consistently outperforms LoRA on fine-tuning tasks like LLaMA, LLaVA, and VL-BART. Tech stack: Python Why it matters: DoRA provides a more efficient and stable fine-tuning approach for large language models, enhancing both the learning capacity and training stability of LoRA without additional inference overhead. It is particularly useful for fine-tuning models like LLaMA, LLaVA, and VL-BART on various downstream tasks such as commonsense reasoning, visual instruction tuning, and image/video-text understanding. Repository: NVlabs/DoRA Primary language: Python GitHub stars: 1003 GitHub description: [ICML2024 (Oral)] Official PyTorch implementation of DoRA: Weight-Decomposed Low-Rank Adaptation [ICML2024 (Oral)]. [](https://github.com/NVlabs/DoRA/stargazers) [

### Overview: This repository is a curated list of awesome awesomeness about artificial intelligence (AI). It includes resources, projects, and papers that showcase the latest advancements in AI, from machine learning to deep learning and beyond. The list is intended for developers and researchers interested in staying updated with the latest developments in AI. ### Key Features: - **Machine Learning (ML)**: Resources and papers for machine learning, including source code and interpretability. - **Deep Learning (DL)**: Papers and resources for deep learning, including architectures and accelerators. - **Computer Vision (CV)**: Materials and resources for computer vision, including image classification, registration, and action recognition. - **Natural Language Processing (NLP)**: Papers and resources for natural language processing, including text classification and image colorization. - **Speech Recognition**: Papers and resources for speech recognition. - **Other Research Topics**: A section dedicated to other research topics in AI, including architecture search and embodied vision. ### Tech Stack: The repository uses Markdown for its structure and includes links to various resources and papers. ### Why It Matters: This curated list is essential for developers and researchers looking to stay updated with the latest developments in AI. It provides a comprehensive overview of the field, making it easier to find relevant resources and papers for specific projects or research needs. ### GitHub Stars: The repository has been starred by 1,004 developers, indicating its popularity and usefulness in the AI community. ### Repository: The primary language of the repository is unknown, but it is hosted on GitHub. The repository

Overview: LangKit is an open-source toolkit designed to monitor Large Language Models (LLMs) by extracting relevant signals from prompts and responses, ensuring safety and security through text quality, relevance, and sentiment analysis. Key features: - Text Quality Metrics - Text Relevance Metrics - Security and Privacy Metrics - Sentiment and Toxicity Analysis Tech stack: Python Why it matters: LangKit provides a comprehensive tool for LLM observability, ensuring the safety and security of language models by monitoring their behavior and extracting meaningful metrics from their interactions. Repository: whylabs/langkit GitHub stars: 1004

Overview: SAGA, or Segment Any 3D GAussians, is an official implementation of a method that extends the capabilities of 3D Gaussian splatting to handle large-scale scenes. This project aims to provide a comprehensive solution for interactive segmentation and rendering of 3D scenes. Key features: - Interactive GUI for real-time segmentation - Support for both 3D Gaussian splatting and open-vocabulary segmentation - Pre-trained 3D Gaussian splatting model for efficient rendering - Support for various datasets including 360_v2, nerf_llff_data, and LERF Tech stack: Jupyter Notebook Why it matters: SAGA offers a powerful tool for interactive 3D scene segmentation, making it ideal for applications in augmented reality, virtual reality, and other fields where real-time rendering and segmentation are required. Its ability to handle large-scale scenes efficiently, thanks to pre-trained models, makes it a valuable addition to the toolkit for 3D rendering and analysis. Repository: Jumpat/SegAnyGAussians GitHub stars: 1002 GitHub description: The official implementation of Segment Any 3D GAussians (AAAI-25)

Overview: This project allows users to generate images from any camera viewpoint using 3D interactive control. It supports both Chinese and English interfaces and can be deployed freely. The system is built with Three.js and Gradio, and it includes features like 8x4 view combinations, real-time preview, and batch generation. Key features: - 3D interactive camera control - 8x4 view combinations - Quick presets for common viewpoints - Real-time preview - Bilingual UI - History gallery for image browsing - Batch generation Tech stack: Python, Gradio, Three.js Why it matters: This project provides a free, transparent, and self-hostable solution for generating images from any camera viewpoint, making it accessible to everyone without the need for a subscription.

### Overview: Neural Kernel Surface Reconstruction (NKSR) is a novel method for reconstructing 3D implicit surfaces from large-scale, sparse, and noisy point clouds. The approach builds upon the recently introduced Neural Kernel Fields (NKF) representation, offering improved generalization capabilities, robustness to noise, and reduced training requirements. NKSR can reconstruct millions of points in a few seconds and handle very large scenes in an out-of-core fashion, achieving state-of-the-art results on various reconstruction benchmarks. ### Key features: - **Compact Kernel Functions**: Enables the use of memory-efficient sparse linear solvers, scaling to large scenes. - **Noise Robustness**: Solves through a gradient fitting approach, making the method robust to noise. - **Flexibility in Training Data**: Can learn from any dataset of dense oriented points, including mixing objects and scenes of different scales. ### Tech stack: Cuda ### Why it matters: NKSR offers a significant improvement in the field of 3D surface reconstruction, making it highly relevant for applications such as computer graphics, robotics, and virtual reality. Its ability to handle large-scale data and noise effectively makes it a powerful tool for various real-world applications.

### Overview: Lawyer LLaMA 是一个通过指令微调的中文法律LLaMA模型，旨在弥补法律领域数据不足的问题，提升模型在法律领域的应用能力。 ### Key features: - **法律知识掌握**：能够正确理解民法、刑法、行政法、诉讼法等常见领域的法律概念。 - **法律知识应用**：以通俗易懂的语言解释法律概念，并进行基础的法律咨询，涵盖婚姻、借贷、海商、刑事等法律领域。 - **指令微调数据**：开源的法律领域指令微调数据和基于LLaMA训练的中文法律大模型参数。 ### Tech stack: Python ### Why it matters: 通过Lawyer LLaMA，我们希望填补法律领域LLaMA模型数据的空白，提升模型在法律领域的应用能力，为法律专业人士和普通用户提供更准确、实用的法律咨询服务。

Overview: Autollm is a powerful tool that simplifies the creation of RAG-based LLM web applications, unifying multiple language models, vector databases, and cost calculations into a single, streamlined solution. Key features: - Supports 100+ language models - Unified API access - 20+ vector databases integration - Cost calculation for 100+ language models - 1-line RAG LLM engine - 1-line FastAPI implementation Tech stack: Python Why it matters: Autollm provides a seamless and efficient way to deploy RAG-based LLM applications, making it easier for developers to create and scale their applications with minimal effort.

Overview: Claude Delegator is a plugin for Claude Code that allows developers to delegate complex tasks to GPT and Gemini specialists, ensuring right specialists for each problem type and auto-routing based on the request. Key features: - Integrates with Codex and Gemini for GPT and Gemini specialists respectively. - Provides 5 domain experts with distinct specialties. - Supports dual mode for analyzing or implementing. - Automatically routes tasks based on the request. - Supports both Codex and Gemini providers. - Synthesizes expert responses to improve clarity. Tech stack: JavaScript Why it matters: Claude Delegator automates the delegation of complex tasks to GPT and Gemini specialists, ensuring the right expertise is used for each problem type, making the development process more efficient and effective. Repository: jarrodwatts/claude-delegator Primary language: JavaScript GitHub stars: 1005 GitHub description: Delegate tasks to Codex and Gemini directly from within Claude Code.

### Overview: Sophia is an optimized optimizer designed specifically for language model pre-training, providing significant improvements in training speed and efficiency. ### Key Features: - **SophiaG Optimizer**: Implements the "Sophia-G" optimizer from the paper [https://arxiv.org/abs/2305.14342](https://arxiv.org/abs/2305.14342). - **Stochastic Second-Order Optimization**: Utilizes stochastic second-order methods for efficient training. - **Scalability**: Designed to handle large language models efficiently. - **Ease of Use**: Based on existing libraries, making integration straightforward. ### Tech Stack: - **Python**: The primary language used for implementation. - **PyTorch**: Used for training and optimization. ### Why It Matters: - **Speed and Efficiency**: Significantly faster training compared to traditional optimizers. - **Scalability**: Optimized for large language models, making it suitable for pre-training tasks. - **Ease of Integration**: Built on existing libraries, making it easy to integrate into existing projects. ### Repository: Liuhong99/Sophia - **Stars**: 1001 - **Description**: Official implementation of "Sophia: A Scalable Stochastic Second-order Optimizer for Language Model Pre-training" and GPT-2 training scripts. - **Language**: Python ### Dependencies: - **PyTorch**: 2.1.2 - **transformers**: 4.33.0 - **datasets** -

Overview: CALVIN is an open-source benchmark designed to evaluate the performance of language-conditioned policy learning for long-horizon robot manipulation tasks. It provides a comprehensive dataset and flexible specification of sensor suites, aiming to enable the development of agents capable of solving complex tasks over extended horizons. Key features: - A comprehensive dataset for long-horizon robot manipulation tasks. - Flexible specification of sensor suites, including vision and language. - Support for various vision and language datasets. - Simulated environment for training and evaluation. - Benchmarked with state-of-the-art methods. Tech stack: Python Why it matters: CALVIN serves as a crucial benchmark for researchers and developers in the field of language-conditioned policy learning, providing a realistic challenge that can accelerate the development of advanced robotic systems capable of handling complex tasks over extended horizons.

Overview: CareGPT is a medical language model that integrates multiple publicly available medical fine-tuned datasets and open-source medical large language models. It aims to accelerate the development of medical LLMs through comprehensive training, evaluation, and deployment. Key features: - Fine-tuning with ChatGPT - Deployment of ChatGPT-Next-Web - Gradio deployment - Training of LLaMA and LLaMA-2 - Support for LoRA and QLoRA - Integration with knowledge base - Open-source over 60 hospital department guide materials - Developed a tool for distilling medical data from GPT-4/ChatGPT models - Comprehensive collection of open-source medical LLMs, training data, deployment resources, and evaluation materials Tech stack: Python Why it matters: CareGPT is a significant contribution to the field of medical language models, aiming to accelerate the development of these models through comprehensive training, evaluation, and deployment. It leverages open-source resources to create a robust and accessible platform for medical LLMs, making it easier for researchers and practitioners to contribute to and benefit from this technology.