Coding & Development

tiny-dnn is a C++14 implementation of deep learning, designed for environments with limited computational resources, such as embedded systems and IoT devices. It stands out as a header-only and dependency-free framework, meaning there's nothing to install beyond a C++14 compiler. This makes it highly portable and easy to integrate into existing applications. The framework supports a variety of network layers, activation functions, loss functions, and optimization algorithms, allowing for the construction of diverse deep learning models. It offers reasonable speed without a GPU, leveraging TBB threading and SSE/AVX vectorization. Additionally, tiny-dnn can import models from Caffe and provides a simple, exception-free operational model, making it a good choice for learning neural networks.

WenetSpeech offers a comprehensive 10000+ hour multi-domain Chinese corpus specifically designed for speech recognition tasks. This extensive dataset is compiled from YouTube and Podcast sources, utilizing both Optical Character Recognition (OCR) and Automatic Speech Recognition (ASR) techniques for labeling. To ensure high quality, the corpus undergoes a novel end-to-end label error detection method for validation and filtering. It categorizes data into High Label, Weak Label, and Unlabel sets, suitable for supervised, semi-supervised, or unsupervised training. The dataset also provides various training subsets (S, M, L) and evaluation sets (DEV, TEST_NET, TEST_MEETING) to support diverse ASR system development and benchmarking. Access to the dataset requires visiting the official website, agreeing to the license, and obtaining a password.

Void Editor is an open-source AI code editor built as a fork of VS Code, providing a familiar environment for developers. It emphasizes full privacy by allowing direct connections to any LLM provider and supporting self-hosting of open-source models like DeepSeek, Llama, and Gemini. Key AI features include Tab autocompletion, inline Quick Edit, and a versatile Chat with Agent and Gather modes. Users can transfer existing VS Code themes, keybinds, and settings seamlessly. Void Editor differentiates itself by not routing messages through a private backend, ensuring users retain full control over their data. It also supports advanced features like Checkpoints for LLM changes, Lint Error Detection, custom FIM model support, and fast application of changes, even on large files.

AgentBench is a comprehensive benchmark designed to evaluate Large Language Models (LLMs) as agents across a diverse spectrum of environments. It encompasses 8 distinct environments, including 5 newly created domains like Operating System (OS), Database (DB), Knowledge Graph (KG), Digital Card Game (DCG), and Lateral Thinking Puzzles (LTP), alongside 3 recompiled from published datasets (House-Holding, Web Shopping, Web Browsing). The platform offers both Dev and Test splits for each dataset, requiring LLMs to generate responses thousands of times for thorough evaluation. AgentBench also introduces VisualAgentBench for evaluating and training visual foundation agents based on large multimodal models (LMMs), covering embodied, GUI, and visual design environments. It supports quick setup using Docker Compose and provides benchmarking results via a leaderboard.

EASYChatGPT is an open-source desktop application project designed to facilitate developer access to ChatGPT. It provides a straightforward way for users to interact with ChatGPT's interface directly from their desktop environment, requiring only a personal API key. The project emphasizes ease of use, with a two-step setup process for installation and conversation initiation. It's particularly useful for developers who want to experiment with ChatGPT functionalities without needing to rely on web interfaces or complex setups. The tool currently supports single-turn conversations and requires users to replace the API key in the configuration file. It's important to note that this is a personal project and not an official OpenAI product.

camel_tools is a comprehensive, open-source Python toolkit developed by the CAMeL Lab at New York University Abu Dhabi, specifically designed for Arabic natural language processing. It offers a wide array of functionalities including text pre-processing, advanced morphological modeling, and specialized components for Dialect Identification, Named Entity Recognition, and Sentiment Analysis. The tool is built to be accessible for researchers and developers, with clear installation instructions for various operating systems like Linux, macOS, and Windows. It also provides options for installing necessary data packages, making it a robust solution for anyone working with the complexities of the Arabic language in NLP tasks.

CLUENER2020 offers a PyTorch implementation of various models for Named Entity Recognition (NER), focusing on Chinese language tasks. It includes baseline code for the CLUENER2020 competition, featuring models like BiLSTM-CRF, BERT-base with Softmax/CRF/BiLSTM+CRF, and Roberta with Softmax/CRF/BiLSTM+CRF. The project utilizes the CLUENER2020 dataset, a Chinese fine-grained NER dataset derived from THUCNEWS, with 10 distinct categories such as organization, person name, and address. Users can configure model parameters and other hyperparameters, and the repository provides instructions for setting up the environment and running the models. It also includes pre-trained BERT and Roberta models for convenience.

Feathr is a scalable, unified data and AI engineering platform widely used in production at LinkedIn and now an open-source project under the LF AI & Data Foundation. It allows users to define data and feature transformations using Pythonic APIs, register these transformations, and share them across teams. Particularly useful for AI modeling, Feathr automatically computes and joins feature transformations to training data with point-in-time correctness to prevent data leakage. It supports materializing and deploying features for online production use, offers native cloud integration with scalable architecture, and has been battle-tested for over six years. Feathr handles billions of rows and petabyte-scale data with built-in optimizations, providing rich transformation APIs including time-based aggregations and sliding window joins. It also features a built-in registry for feature reuse and an intuitive UI for searching and exploring features and their lineages.

Legends-Of-Heroes is an open-source C# game framework built on ET 8.1, designed for dual-end game development using .NET8 and Unity6000. It features a comprehensive battle system with ECS skills and a buff system, with ongoing development for skill and behavior tree editors. The framework includes a LOL-style ball battle demo that utilizes state synchronization, where all collision detection, skills, and AI logic are executed server-side. It integrates with EUI, Luban, and YooAsset for UI, configuration, and resource management, respectively. The project also supports HybridCLR for hot updates and offers one-click packaging for various platforms like Windows and Android. It's ideal for developers looking for a robust foundation for real-time multiplayer games with advanced combat mechanics.

Rivet.dev provides infrastructure for software that thinks, offering composable stateful compute for AI agents, collaborative applications, and durable execution. It features 'Actors' as a primitive for stateful workloads, providing in-memory state, indefinite runtime with idle hibernation, infinite scalability, and a global edge network. The platform also includes 'agentOS', a lightweight open-source operating system for agents built on WASM & V8, boasting near-zero cold start times and cost efficiency. Rivet.dev supports various infrastructures like serverless, containers, or on-premise deployments, and integrates with popular frameworks and runtimes. It also offers built-in observability tools for debugging and monitoring actors and agents from local development to production.

OMG is an advanced open-source framework designed for occlusion-friendly personalized multi-concept generation within diffusion models, as presented at ECCV 2024. It allows users to generate complex images featuring multiple characters and styles, integrating seamlessly with LoRAs from Civitai.com and InstantID for single-image ID personalization. The tool also supports ControlNet for layout control and various style LoRAs. OMG is built on Python 3.10.6 with PyTorch 2.0.1 and torchvision 0.15.2, requiring specific model downloads for its functionality, including Stable Diffusion XL and various ControlNet and LoRA checkpoints. It offers flexible usage through command-line inference scripts for both LoRA and InstantID workflows.

unet.cu is an open-source project that provides a UNet diffusion model implemented entirely in C++/CUDA. Inspired by Andrej Karpathy's llm.c, the goal is to achieve performance comparable to PyTorch implementations, specifically for training unconditional diffusion models. The repository includes benchmarks showing its training speed relative to PyTorch and PyTorch with `torch.compile`. It supports training with sample images from ImageNet 64x64 and allows users to train with their own data. The project emphasizes learning CUDA concepts and provides a detailed breakdown of its architecture, including custom convolution kernels and optimizations to avoid inefficient data transposes.

proxylessnas is an open-source tool designed for direct neural architecture search, enabling efficient optimization of deep learning models on target tasks and hardware. It eliminates the need for proxy tasks, directly searching for optimal architectures. The tool is integrated into popular platforms like PytorchHub, Microsoft NNI, and Amazon AutoGluon, making it accessible for various development environments. Notably, proxylessnas achieved first place in the Visual Wake Words Challenge at CVPR 2019, demonstrating its effectiveness in specialized applications. It supports specialization of architectures for different platforms, such as CPU, GPU, and mobile devices, to fully exploit efficiency.

antigravity-agent is an open-source tool designed for the effortless management of multiple Antigravity accounts. It allows users to quickly switch between accounts with a single click, eliminating the need for repetitive logins. The software automatically identifies and saves current account data, and offers secure backup functionality through password-encrypted export of account configurations, facilitating cross-device migration. A VSCode extension is also available, enabling users to switch accounts and view model quotas directly within their editor. The tool emphasizes security, distributing only through official GitHub Releases to protect sensitive account information.

Stellon Labs is an AI research lab dedicated to developing powerful, tiny AI models specifically optimized for edge applications. Their focus is on creating 'frontier AI' solutions that can operate efficiently on minimal hardware, making advanced artificial intelligence accessible for devices with limited computational resources. The lab aims to push the boundaries of AI performance in constrained environments, enabling new possibilities for on-device intelligence without requiring extensive infrastructure. Their work is geared towards practical applications where low-power and small-footprint AI is crucial.

DenoisingDiffusionProbabilityModel-ddpm- is an open-source implementation of the Denoising Diffusion Probability Model (DDPM). This tool provides a straightforward way for developers and researchers to train a UNet model on the CIFAR-10 dataset. Users can directly run `Main.py` to initiate training and then adjust model configurations to visualize the denoising process. The repository also includes `MainCondition.py` for training with Classifier-free guidance. Pre-trained weights for CIFAR-10 are available, and the project references key papers and blogs for deeper understanding of DDPM frameworks, making it an accessible resource for learning and experimentation in diffusion models.

JittorLLMs is a large language model inference library designed for high performance, low configuration requirements, and excellent Chinese language support. A key differentiator is its ability to run large models on machines with as little as 2GB RAM and no dedicated GPU, making local deployment accessible to a wider range of users. The library supports a variety of popular models including ChatGLM, Peng Cheng PanGu, ChatRWKV, LLaMA/LLaMA2, MOSS, and Atom7B, with plans to integrate more domestic models. It boasts significant speed improvements, reducing model loading times by 40% and boosting computational performance by over 20% compared to similar frameworks, thanks to zero-copy technology and automatic meta-operator compilation. JittorLLMs also offers high portability, allowing users to migrate models to various heterogeneous computing devices and environments by simply installing Jittor-version Torch (JTorch) without code modification. It includes features for memory optimization, such as dynamic swap technology, to manage memory and VRAM usage efficiently.

my-neuro is an open-source project designed to help users create their own personalized AI desktop companions. Inspired by Neuro Sama, this tool allows for extensive customization of characters, including voice, personality, and appearance, compatible with various Live2D models. It boasts ultra-low latency responses, with conversations responding in under one second, and supports both local inference with open-source LLMs and integration with closed-source AI models via DMXAPI. Key features include long-term memory, visual recognition, voice cloning, and LLM training, enabling the AI to remember user interactions, understand visual cues, and adapt its responses. The project also plans to integrate advanced human-like interaction designs, such as real-time interruptions, emotional responses, and desktop control capabilities, making it a versatile platform for building deeply personal AI companions.

MLDB is an open-source SQL database specifically engineered for machine learning applications. Developed by MLDB.ai, it allows users to install it as a command-line tool, run scripts, or interact via a RESTful API. Key functionalities include storing data, exploring it using a specialized SQL dialect, training machine learning models, and deploying these models as APIs. The database is designed for high efficiency in data loading, classical ML algorithm training, and prediction endpoints. It features a data model and type system optimized for ML, supporting nested structures, embeddings, and tensors. MLDB is extensible through C++, Python, and Javascript plugins, and is currently being rearchitected for a smaller core and broader deployment platforms, aiming to simplify the creation and deployment of ML solutions.

retina-unet is an open-source convolutional neural network specifically designed for the segmentation of blood vessels in retina fundus images. Based on the U-Net architecture, this tool performs a binary classification task, identifying each pixel as either a vessel or not. It has been rigorously tested on the DRIVE and STARE databases, demonstrating superior performance in terms of area under the ROC curve compared to other methods. The repository provides the implementation in Python, utilizing the Keras library with either Theano or TensorFlow backends. It includes detailed instructions for data preparation, training with sub-images (patches), and evaluating the trained model, making it a valuable resource for medical image analysis and research.

Pytorch-bert-crf-ner offers a PyTorch implementation for Korean Named Entity Recognition (NER) tagging, leveraging the power of BERT and CRF models. This open-source tool is specifically designed to assist in Korean Natural Language Processing (NLP) tasks and research. It provides functionalities to identify and classify named entities such as persons, locations, organizations, dates, and more within Korean text. The repository includes examples, data utilities, and training scripts, making it suitable for developers and researchers working with Korean language data who need to implement or experiment with NER models.

Pytorch-maddpg offers a PyTorch implementation of the Multi-Agent Deep Deterministic Policy Gradient (MADDPG) algorithm, a key approach in multi-agent reinforcement learning. This open-source project is hosted on GitHub and is designed for researchers and developers working on complex multi-agent systems. The implementation includes a modified Waterworld environment, where agents (evaders, pursuers, poisons) interact under specific physical rules, allowing for experimentation with cooperative behaviors. It supports features like agents bouncing off walls and requiring exact cooperation for rewards, making it a valuable tool for studying multi-agent coordination and policy learning.

XVERSE-13B is a multilingual large language model developed by XVERSE Technology Inc. It features a Decoder-only Transformer network structure with an 8K context length, which is extended to 256K in the XVERSE-13B-256K version for handling extensive input content like literature summaries and report analysis. The model was trained on 3.2 trillion tokens across over 40 languages, with a focus on Chinese and English performance. It utilizes a 100,534-token BPE-based tokenizer that supports multiple languages without requiring additional vocabulary expansion. The project also highlights an efficient training framework with high peak computing power utilization. Quantized models (GGUF, GPTQ) are available for inference on MacOS, Linux, and Windows systems.

SwanLab is an open-source, modern-design AI training tracking and visualization tool built for AI model training teams. It provides comprehensive features for experiment analysis, metric observation, and collaboration. Researchers can track key metrics, record hyperparameters, and visualize training processes through an intuitive UI, helping to identify issues and accelerate model iteration. SwanLab supports a wide range of data types including scalar metrics, images, audio, text, video, 3D point clouds, and biochemical molecules, along with various chart types like line, media, bar, and custom ECharts. It offers both cloud and self-hosted deployment options and integrates with over 50 mainstream frameworks, including PyTorch, Transformers, and Keras. Key functionalities include experiment comparison, multi-person collaboration, hardware monitoring, and an open API for extended capabilities.