Coding & Development

stable-fast is an ultra-lightweight inference optimization framework specifically designed for HuggingFace Diffusers on NVIDIA GPUs. It achieves state-of-the-art inference performance across various diffuser models, including StableVideoDiffusionPipeline, with compilation times of only a few seconds, unlike other solutions that can take dozens of minutes. The framework supports dynamic shapes, LoRA, and ControlNet, and integrates key techniques such as CUDNN Convolution Fusion, Low Precision & Fused GEMM, Fused Linear GEGLU, NHWC & Fused GroupNorm, and CUDA Graph. It also improves the `torch.jit.trace` interface for more stable tracing of complex models and offers dynamic quantization for VRAM reduction, making it a powerful tool for developers working with AI models.

SUPIR is an open-source project dedicated to developing practical algorithms for photo-realistic image restoration in real-world scenarios. It provides advanced capabilities for enhancing image quality, including super-resolution and the ability to handle various degradations. The project emphasizes achieving high generalization and image quality, with options for both quality-oriented and fidelity-oriented settings. Users can choose between different model versions (SUPIR-v0Q and SUPIR-v0F) depending on their specific needs, such as general high quality or better detail preservation for light degradations. An online demo, SupPixel AI, is also available for easy access to its cutting-edge AI technology for image processing and upscaling.

StockPredictionRNN is an open-source project designed for high-frequency trading price prediction, leveraging LSTM Recursive Neural Networks. This tool is specifically engineered to forecast prices within high-frequency stock exchange environments. It implements its prediction solution using historical data from NYSE OpenBook, allowing users to recreate the limit order book for any given time. The project is written in Python 2.7 and utilizes the Keras library, along with dependencies like Theano, numpy, scipy, matplotlib, and pymongo. It provides instructions for data acquisition from NYSE FTP servers and a clear installation and usage guide for setting up the environment and running the prediction models.

RSL Solution provides pre-vetted remote developers across various specializations including AI, Python, Hardware Design, and Full Stack. With a talent pool of over 5,000 expert developers available in countries like India, USA, UK, Canada, Barbados, and Ghana, RSL aims to help companies scale their teams rapidly. They offer a 48-hour deployment promise, ensuring that businesses can onboard skilled professionals quickly. The service includes rigorous vetting, background checks, and flexible engagement models such as project-based, hourly, or dedicated teams. RSL emphasizes quality assurance with a 99% success rate, continuous monitoring, and performance guarantees, making it suitable for businesses looking for reliable tech talent.

SupContrast offers a PyTorch implementation of "Supervised Contrastive Learning" and, incidentally, "A Simple Framework for Contrastive Learning of Visual Representations" (SimCLR). This repository serves as a reference, illustrating these methods using CIFAR datasets. It includes a `SupConLoss` function that takes features and labels, degenerating to SimCLR loss if labels are not provided. The implementation provides comparison results on CIFAR-10 and CIFAR-100, showcasing improved accuracy over standard cross-entropy. It also details running instructions for standard cross-entropy, supervised contrastive learning, and SimCLR, including pretraining and linear evaluation stages, and supports custom datasets.

Teachable Machine Community is an open-source repository offering example code snippets and machine learning code for Teachable Machine. Teachable Machine is a web-based tool designed to make machine learning model creation fast, easy, and accessible for everyone, including educators, artists, students, and innovators. Users can train a computer to recognize images, sounds, and poses without needing prior machine learning knowledge or coding. The repository includes a libraries section with machine learning code utilizing Tensorflow.js for in-browser model training and execution, along with API helper libraries for integrating exported models into projects. It also features a snippets section with code and instructions for using Teachable Machine models in languages like Javascript, Java, and Python.

Distribution-is-all-you-need is an open-source GitHub repository offering a comprehensive tutorial on fundamental probability distributions crucial for deep learning researchers. The resource leverages Python libraries to illustrate various distributions, including Uniform, Bernoulli, Binomial, Categorical, Multinomial, Beta, Dirichlet, Gamma, Exponential, Gaussian, Normal, Chi-squared, and Student-t. It delves into concepts like conjugate distributions and their relevance in Bayesian probability theory, explaining how prior and posterior distributions relate. The tutorial provides code examples for each distribution, making it a practical guide for understanding the mathematical underpinnings of deep learning models.

USearch is a fast, open-source search and clustering engine designed for vectors and arbitrary objects. It offers a highly optimized HNSW implementation, boasting up to 10x faster performance than FAISS. The engine supports a wide array of programming languages including C++, Python, JavaScript, Rust, Java, Objective-C, Swift, C#, GoLang, and Wolfram, making it broadly compatible across different development environments. Key features include SIMD-optimized and user-defined metrics with JIT compilation, hardware-agnostic half-precision support (bf16, e5m2, i8), and the ability to view large indexes from disk without loading them into RAM. USearch also provides heterogeneous lookups, on-the-fly deletions, and binary Tanimoto/Sorensen coefficients for specialized applications like genomics. Its compact codebase and native bindings contribute to lower call latencies and faster deployments.

Zygote AI is an advanced platform designed to facilitate the foundational development of artificial intelligence components and systems. It empowers users to generate initial AI models and prototypes, streamlining the early stages of AI project creation. The platform focuses on providing the necessary tools and environment for developers and researchers to build and experiment with AI models from the ground up. While specific features are not detailed on the provided homepage, the core offering revolves around supporting the creation and iteration of AI components, making it a valuable resource for those involved in AI development and research.

Mintplex Labs Inc is an AI tool development company dedicated to creating innovative AI solutions for a wide range of business applications. Their core mission is to empower businesses by providing custom AI tools designed to enhance efficiency, streamline operations, and improve overall processes. While specific features are not detailed on their public-facing pages, their focus is on developing and deploying AI technologies that can be tailored to meet unique organizational needs. This suggests a strong emphasis on custom development and integration, rather than off-the-shelf products, making them a potential partner for businesses looking to leverage AI for competitive advantage.

TextFooler is an open-source model designed for natural language attack on text classification and inference tasks. It provides the source code and datasets necessary to reproduce research findings related to adversarial attacks on NLP models. This tool is particularly useful for evaluating the robustness of models like BERT, LSTM, and CNN against various adversarial techniques. Researchers and developers can use TextFooler to generate adversaries for text classification and natural language inference, helping them understand and improve the security of their NLP systems. The repository includes detailed instructions for setup, prerequisites, and running attack simulations, making it a valuable resource for adversarial NLP research.

tidybot2 is an open-source project providing a holonomic mobile manipulator designed for robot learning. It includes comprehensive hardware designs and software components for building and operating the robot. The platform supports various tasks, from phone teleoperation and data collection to policy training and inference. Its holonomic base allows for independent and simultaneous control of planar degrees of freedom, simplifying complex mobile manipulation tasks. The project offers a simulation environment for testing the codebase without physical hardware and detailed guides for assembly, usage, and software setup, making it accessible for researchers and developers in the field of robotics.

Time-Series-Forecasting-and-Deep-Learning is a comprehensive, open-source GitHub repository dedicated to curating resources for time series forecasting and deep learning. It serves as a valuable hub for researchers, data scientists, and students seeking to explore the latest advancements in the field. The repository meticulously organizes research papers, including those from 2017 up to 2026, alongside benchmarks, applications like TimeGPT, and various datasets. Additionally, it provides links to relevant courses, blogs, and code libraries, making it an all-in-one reference for anyone involved in time series analysis and model development. The structured content, including a table of contents, allows for easy navigation through a vast collection of academic and practical materials.

tf2_course is a comprehensive collection of Jupyter notebooks designed to accompany the "Deep Learning with TensorFlow 2 and Keras" training. This open-source project, available on GitHub, provides practical exercises and their corresponding solutions, making it an invaluable resource for individuals looking to deepen their understanding and skills in deep learning using TensorFlow 2 and Keras. Users can access these notebooks online via services like Colaboratory, Binder, or Deepnote for temporary environments, or install them locally for a persistent setup. The project also includes detailed installation instructions and addresses common issues like Python version compatibility and SSL errors, ensuring a smooth learning experience for students and professionals alike.

Text-Classification is an open-source project that provides implementations of several state-of-the-art text classification models using TensorFlow. It supports various models including Attention is All You Need, IndRNN, Attention-Based Bidirectional LSTM, Hierarchical Attention Networks, Adversarial Training Methods, Convolutional Neural Networks, and RMDL. The tool is designed for developers and researchers working on text classification tasks, particularly on datasets like DBpedia. It requires Python 3 and TensorFlow 1.4 or later, with updated code for preprocessing using `tf.keras.preprocessing.text`. The repository also includes performance metrics for each implemented model, offering a valuable resource for comparing different approaches.

tslearn is an open-source machine learning toolkit specifically designed for time series analysis in Python. It provides a wide array of functionalities for tasks such as clustering, classification, and regression of time series data. The toolkit supports various data preprocessing steps, including scaling and resampling, and offers different distance metrics like Dynamic Time Warping (DTW). tslearn is built to be compatible with scikit-learn's API, allowing users to leverage familiar utilities for hyper-parameter tuning and pipelines. It also includes features for calculating barycenters, performing early classification, and working with UCR datasets, making it a versatile tool for researchers and practitioners in the field.

SoTA-Point-Cloud is a GitHub repository offering an extensive survey of deep learning techniques applied to 3D point clouds. Published in IEEE TPAMI 2020, this resource covers major tasks such as 3D shape classification, 3D object detection, and 3D point cloud segmentation. It provides comparative results on numerous publicly available datasets, including ModelNet, KITTI, and Semantic3D. The repository also offers insightful observations and outlines future research directions, making it an invaluable resource for researchers and practitioners in the field of 3D computer vision. The maintainers regularly update the page with new results and suggestions.

UAV_Obstacle_Avoiding_DRL is a comprehensive open-source project focused on developing deep reinforcement learning algorithms for autonomous obstacle avoidance in Unmanned Aerial Vehicles (UAVs). It addresses both static and dynamic environments, offering multiple approaches for each. For static environments, the project explores Multi-Agent Reinforcement Learning (MADDPG, DDPG, TD3) combined with artificial potential field algorithms. In dynamic settings, it utilizes disturbed flow field algorithms alongside single-agent reinforcement learning (PPO+GAE, TD3, DDPG, SAC). The project also includes implementations of traditional path planning methods like A* search, RRT, Ant Colony Algorithm, and D* algorithm for comparison, highlighting the superior performance of reinforcement learning approaches. It provides both MATLAB and Python implementations for various algorithms, making it a valuable resource for researchers and developers in UAV navigation.

This project implements a Stock Trading Bot utilizing Deep Reinforcement Learning, specifically Deep Q-learning. It's designed for learning and experimentation, keeping the implementation simple and close to the algorithm discussed in research papers. The bot allows users to create intelligent agents that learn from market data, making decisions to buy, sell, or hold based on observed states. It incorporates several improvements to the Q-learning algorithm, including Vanilla DQN, DQN with fixed target distribution, Double DQN, Prioritized Experience Replay, and Dueling Network Architectures. Users can train the agent on historical data and evaluate its performance, with visualizations available for model evaluations. It's a valuable resource for those interested in applying reinforcement learning to financial trading.

zh-NER-TF is an open-source project offering a straightforward character-based BiLSTM-CRF model specifically designed for Chinese Named Entity Recognition (NER). This TensorFlow-based tool aims to identify three key entity types: PERSON, LOCATION, and ORGANIZATION within Chinese text. The model utilizes a look-up layer for character embeddings, a BiLSTM layer to extract features from both past and future input, and a CRF layer to ensure grammatically correct tag sequences, addressing limitations of simpler Softmax layers. It includes preprocessed data files and a vocabulary for easy setup, and users can train, test, or demo the model with their own datasets after transforming them into the specified format. The repository provides instructions for running the model and evaluating its performance.

xLearn is a robust, high-performance machine learning package developed in C++ for maximum CPU and memory utilization. It includes implementations of linear models (LR), factorization machines (FM), and field-aware factorization machines (FFM), making it ideal for solving large-scale machine learning problems, particularly with high-dimensional sparse data common in recommendation systems. The package is designed for ease of use, requiring no third-party libraries for compilation and offering simple Python and CLI interfaces. xLearn also boasts scalability, supporting out-of-core training to handle terabytes of data by leveraging disk storage, and includes features like cross-validation and early-stop mechanisms.

The Yi series models are a collection of open-source large language models developed from scratch by 01.AI. These models are designed to be bilingual, trained on a 3T multilingual corpus, and excel in language understanding, commonsense reasoning, and reading comprehension. The Yi-34B-Chat model has demonstrated strong performance, ranking highly on leaderboards like AlpacaEval. The series includes both chat-optimized and base models, with options for different parameter sizes (6B, 9B, 34B) and context window lengths (up to 200K). Yi models are built on the Transformer architecture, similar to Llama, but are not derivatives, utilizing independently created training datasets and infrastructure. They are available for deployment via pip, Docker, conda-lock, and llama.cpp, and can be fine-tuned or quantized for specific needs.

ZynqNet is an open-source project stemming from a Master Thesis, focusing on FPGA-accelerated embedded convolutional neural networks. It provides a comprehensive solution for image classification on embedded systems, featuring the ZynqNet CNN, an optimized and customized CNN topology, and the ZynqNet FPGA Accelerator, an FPGA-based architecture for its evaluation. The project also includes the Netscope CNN Analyzer, a custom tool for visualizing, analyzing, and editing CNN topologies. ZynqNet is designed for high efficiency, achieving 84.5% top-5 accuracy with minimal computational complexity, making it ideal for real-time and power-constrained applications. The repository offers the full project report, CNN prototxt, pretrained weights, HLS C++ source code for the accelerator, and firmware for the Zynq XC-7Z045 ARM processors.

temperature_scaling is an open-source Python module designed to calibrate neural networks by adjusting their confidence scores. Originally created as a demonstration for PyTorch 0.3, it implements temperature scaling, a post-processing technique that divides logits by a learned scalar parameter to minimize negative log-likelihood on a validation set. This helps address the common issue of neural networks outputting overconfident probabilities, ensuring that confidence scores better match true correctness likelihood. While the repository is unmaintained, it offers a clear example of how to integrate temperature scaling into a project for improved model calibration.