Awesome GitHub RepositoriesParameter Sampling

Hyperfine is a command-line benchmarking tool used to measure the execution time of shell commands through multiple runs and statistical analysis. It functions as a comparative benchmarking utility and a shell performance analyzer, allowing for the evaluation of multiple commands against a reference baseline to determine relative speed. The tool distinguishes itself by isolating actual command performance through shell overhead correction and the ability to bypass the shell entirely using system calls. It supports parameterized execution, enabling benchmarks to run across a range of varying i

Albumentations is a computer vision image augmentation library designed to increase training data diversity for deep learning models. It provides a toolset for applying geometric and color transformations to images and annotations, including a specialized collection of 3D operations for volumetric data used in medical and scientific imaging. The library functions as an image mask and bounding box transformer, automatically updating masks, bounding boxes, and keypoints when images undergo geometric changes. This ensures that spatial alterations remain synchronized across images and their assoc

Albumentations is an image augmentation library and computer vision preprocessing tool designed to expand datasets for deep learning models. It provides a collection of transformations that modify pixel values and spatial geometry to increase the diversity of training samples and improve model generalization. The library supports both 2D image augmentation and 3D volumetric data augmentation. It handles a variety of labels alongside images, ensuring that bounding boxes, keypoints, and segmentation masks remain accurately aligned when spatial transformations are applied. The tool incorporates

imgaug is a Python library for machine learning data augmentation and computer vision dataset expansion. It provides tools to increase the volume and variety of training sets by applying random geometric, color, and noise transformations to images. The library ensures spatial consistency by synchronizing transformations across images and their associated annotations, such as bounding boxes, keypoints, and segmentation maps. It uses a compositional pipeline pattern to chain multiple augmentations into sequences and employs deterministic seed management to reproduce specific data samples. The

Optuna is a Python-based hyperparameter optimization framework designed to automate the search for optimal machine learning model configurations. It functions as a Bayesian optimization library that systematically tests parameter combinations to maximize or minimize objective functions, streamlining the model development process through iterative evaluation. The project distinguishes itself through a define-by-run dynamic construction model, which allows users to build complex, conditional search spaces using standard programming logic. Its architecture is highly modular, featuring a pluggabl

MiniCPM is a collection of small language models designed for local, on-device deployment in resource-constrained environments. The project focuses on running dense Transformer models on consumer hardware, including GPUs, CPUs, and Apple Silicon, without requiring custom code forks. The project distinguishes itself through heavy optimization for edge hardware, utilizing quantized weight compression in GGUF and MLX formats to reduce memory overhead. It implements advanced inference techniques such as speculative sampling and radix-tree prefix caching to accelerate generation speed and throughp

This project is a terminal-based command line interface client and agent orchestrator for interacting with multiple large language model providers. It functions as an OpenAI API client and a local API gateway that exposes chat completions and embeddings through an HTTP server. The system distinguishes itself by providing a retrieval-augmented generation tool for indexing local files and URLs into a vector database to provide custom document context. It allows for the creation of specialized AI agents that combine custom system prompts with tool calling and external function execution. The to

The inspector is a diagnostic and validation tool for the Model Context Protocol. It provides an interactive interface and a transport proxy to discover, inspect, and execute the tools, prompts, and resources provided by an MCP server. The project serves as a debugger and compliance tester to verify that server implementations adhere to the protocol specification and JSON-RPC standards. It allows for real-time monitoring of message exchanges and logs between clients and servers across various transport layers, such as standard input/output and Server-Sent Events. The tool covers a broad rang

BetterChatGPT is a cross-platform user interface and OpenAI API client designed for interacting with large language models. It functions as a prompt engineering workspace and a self-hosted AI frontend that allows users to connect to models via API keys or custom proxy endpoints. The project distinguishes itself through conversation management tools, including the ability to organize chats into color-coded folders and maintain a library of reusable prompt templates. It also includes a real-time cost monitoring system that tracks token consumption and calculates estimated pricing for interactio

G0DM0D3 is a static web client and multi-model chat gateway designed for AI research, prompt optimization, and red teaming. It provides a unified interface to query numerous AI models in parallel, allowing for the simultaneous evaluation of different prompt variations and sampling parameters to identify the most successful outputs. The project features specialized tooling for probing safety filters and bypassing model constraints through an input perturbation engine that applies text obfuscation and character substitution. It includes a composite scoring system to rank model performance and a

Hyperopt is a Python library for hyperparameter optimization designed to minimize scalar-valued objective functions. It operates as a stochastic search space engine that finds optimal input parameters by searching through real-valued, discrete, and conditional spaces. The framework distinguishes itself through its support for complex search space configurations, allowing for conditional parameter hierarchies where specific hyperparameters are sampled only if their parent parameters meet certain criteria. It is built as an asynchronous optimization framework, decoupling the generation of searc

Corenet is a deep learning training framework and computer vision model library designed for developing neural networks across vision, text, and audio modalities. It functions as a distributed training orchestrator for scaling workloads across multiple compute nodes and provides a multimodal data pipeline for processing image, text, and video data. The project includes a model conversion toolkit for transforming weights and architectures between different machine learning frameworks. It also provides tools for optimizing model performance on Apple Silicon and reducing response latency in gene

MMDetection3D is an open-source toolbox for 3D perception, providing a unified framework for detecting and segmenting objects in three-dimensional environments. It supports a range of core tasks including monocular 3D object detection from single camera images, LiDAR-based 3D object detection from raw point clouds, and multi-modal fusion that combines camera images with LiDAR data. The toolbox also covers point cloud semantic segmentation, assigning class labels to every point in a scan for scene understanding. The project distinguishes itself through a config-driven pipeline that orchestrate

Arjun is an HTTP parameter discovery tool that identifies valid parameters on web endpoints by testing large dictionaries of parameter names against target URLs. It systematically probes endpoints using GET, POST, JSON, and XML request formats to find which parameters the server accepts, and can detect parameters whose values appear reflected in the response body. The tool distinguishes itself through its multi-method scanning approach, passive parameter collection from public archives like OTX and CommonCrawl, and its ability to detect value-sensitive parameters that only trigger a response

RF-DETR is a Python library for training and deploying object detection, instance segmentation, and keypoint detection models built on a vision transformer architecture. It provides a unified command-line interface and Python API for the full workflow, from fine-tuning pretrained checkpoints on custom datasets to running inference on images, video files, and live camera streams. The project supports training on datasets in COCO or YOLO format, with automatic format detection and configurable augmentation pipelines. Models can be exported to ONNX, TFLite, or TensorRT for deployment across edge

Composer es un framework de entrenamiento distribuido para PyTorch diseñado para escalar modelos a gran escala en clústeres de GPU multi-nodo. Funciona como un entrenador de modelos de lenguaje de gran tamaño (LLM), un optimizador de modelos distribuidos y un gestor del ciclo de vida de entrenamiento. El proyecto se diferencia como una biblioteca de regularización para deep learning, proporcionando técnicas de optimización especializadas como Sharpness Aware Minimization, MixUp y CutMix para mejorar la generalización del modelo. Además, distingue su flujo de entrenamiento mediante el uso de warmup de longitud de secuencia, congelación progresiva de capas y checkpointing de estado fragmentado (sharded-state) para la recuperación de modelos a gran escala. El framework cubre una amplia superficie de capacidades, incluyendo la orquestación de entrenamiento distribuido, la gestión de hardware de precisión mixta y el streaming de datos cloud-native. También proporciona herramientas extensas de monitoreo y observabilidad para diagnósticos de memoria de GPU, detección de divergencia en el entrenamiento y seguimiento del rendimiento (throughput). El proyecto incluye un lanzador de línea de comandos para automatizar la ejecución de trabajos de entrenamiento multi-GPU entre nodos.

AugLy es una biblioteca de aumento de datos multimodal y aumentador de conjuntos de datos de machine learning. Proporciona un sistema para generar variaciones sintéticas de datos de entrenamiento a través de conjuntos de datos de audio, imagen, texto y video para aumentar la diversidad de muestras y mejorar la robustez del modelo. La biblioteca funciona como un simulador de ruido multimedia, diseñado específicamente para imitar capturas de usuarios del mundo real superponiendo plantillas de redes sociales y artefactos de internet sobre los medios. Incluye un rastreador de procedencia de datos para registrar las transformaciones específicas y los niveles de intensidad aplicados a cada pieza de datos aumentados. La herramienta cubre una amplia gama de capacidades de expansión de conjuntos de datos, incluyendo transformaciones lingüísticas para texto, transformaciones temporales y visuales para video, y transformaciones sónicas para audio.

Este repositorio es un programa educativo integral y un framework de deep learning diseñado para enseñar aprendizaje profundo práctico usando PyTorch a través de notebooks y ejemplos de código. Sirve como una librería de alto nivel para construir, entrenar y desplegar redes neuronales, actuando como un orquestador de entrenamiento de modelos que coordina modelos de PyTorch, optimizadores y funciones de pérdida. El proyecto proporciona kits de herramientas especializados para visión artificial, procesamiento de lenguaje natural y preprocesamiento de datos tabulares. Se distingue por controles de entrenamiento avanzados como tasas de aprendizaje discriminativas, un sistema de callbacks bidireccional para personalizar la lógica de entrenamiento y una abstracción de learner de alto nivel que automatiza la colocación en dispositivos y los bucles de entrenamiento. El framework cubre una amplia superficie de capacidades, incluyendo la construcción automatizada de pipelines de datos, análisis de arquitectura de modelos y evaluación de rendimiento en tareas de clasificación, regresión y segmentación. También incluye utilidades para entrenamiento distribuido en múltiples GPUs, entrenamiento de precisión mixta para optimización de memoria y soporte especializado para datos de imágenes médicas. El proyecto se entrega como una serie de Jupyter Notebooks.

Chinese-Vicuna es un modelo de lenguaje grande chino e IA que sigue instrucciones basado en la arquitectura LLaMA. Está diseñado específicamente para la comprensión y generación de lenguaje natural en el idioma chino, utilizando un modelo ajustado por instrucciones para seguir prompts complejos de usuario a través de conversaciones. El proyecto proporciona un framework de ajuste fino LoRA y sistemas de cuantización para permitir la adaptación del modelo y la inferencia en hardware de consumo. Implementa inferencia cuantizada para reducir el uso de memoria tanto en CPUs como en GPUs, soportado por una implementación de bajo nivel en C++ para minimizar los requisitos de recursos del sistema. El sistema cubre una amplia gama de capacidades de procesamiento de lenguaje natural, incluyendo gestión de conversaciones multivuelta, traducción multilingüe y generación de código de programación. También incluye herramientas para entrenamiento específico de dominio, conversión de formato de modelo y una interfaz de chat interactiva con salida de texto en streaming.