Projects sharing features with Xformers

This project is an educational resource and comprehensive guide for implementing and deploying deep learning models using the PyTorch framework. It provides a structured learning curriculum consisting of tutorials and notebooks that cover neural network architectures, data pipelines, and model optimization across multiple AI domains. The curriculum includes practical implementation guides for building convolutional networks, transformers, and recurrent models. It specifically focuses on workflows for computer vision, including image classification, object detection, and segmentation, as well

DeepSpeed is a high-performance library designed to scale deep learning model training and inference across massive clusters of GPUs and compute nodes. It provides a comprehensive suite of tools for distributed training, enabling the execution of models that exceed the memory capacity of single devices through advanced parameter partitioning, pipeline-based model parallelism, and memory-efficient state offloading. The framework distinguishes itself through specialized communication-efficient optimizers and hardware-aware acceleration techniques. By utilizing gradient compression, quantization

Flax is a deep learning framework and JAX neural network library designed for building complex machine learning models. It functions as a distributed training library and model state manager, providing a toolkit for defining flexible neural network architectures and scaling their training across multiple hardware devices. The project is characterized by a design that separates network logic from parameter values to remain compatible with pure functions. It uses hierarchical module composition to organize networks as trees of nested modules and employs a reference-based state management system

AITemplate is an ahead-of-time deep learning compiler that translates PyTorch neural networks into standalone C++ source code. It functions as a PyTorch to C++ compiler and a GPU kernel fusion engine, producing self-contained executable binaries that run inference without requiring a Python interpreter or deep learning framework runtime. The project generates optimized CUDA and HIP C++ code specifically for NVIDIA TensorCores and AMD MatrixCores. It focuses on maximizing throughput for half-precision floating-point operations through a system that combines multiple neural network operators in

Nixtla is a time series analysis platform centered on a transformer-based foundation model. It provides zero-shot inference for forecasting and anomaly detection, allowing the system to predict future values for new time series without requiring model retraining. The project is designed for large-scale analysis, using distributed inference scaling and forecast parallelization to process millions of data series. It supports fine-tuning adaptation to adjust pretrained weights for domain-specific datasets and offers deployment options ranging from local execution and private containers to integr

Apex is a high-performance toolkit for PyTorch designed to coordinate distributed training, execute fused GPU kernels, manage mixed precision, and implement optimized distributed optimizers. It provides specialized tools for scaling model training across multiple GPUs and nodes to increase processing speed and throughput. The library features high-performance implementations of Adam and LAMB optimizers to reduce synchronization overhead and memory bottlenecks. It utilizes fused CUDA kernels to combine neural network operations, reducing memory overhead and increasing execution speed. The too

Flash Linear Attention is a training framework and inference engine for sequence models that use linear attention and state space mechanisms, designed to process long contexts with reduced memory and compute overhead. It provides hardware-optimized token mixing layers and fused CUDA kernels that minimize memory bandwidth and launch overhead across different GPU architectures, and includes a causal inference engine that generates text token-by-token using cached hidden states for efficient autoregressive decoding. The project supports building hybrid sequence models that interleave standard at

WebGPT is a browser-based machine learning framework designed to execute transformer models entirely within the client environment. By leveraging native web standards, it provides a zero-dependency runtime that enables local text generation without the need for backend server processing. The engine distinguishes itself by utilizing hardware-accelerated compute shaders to perform high-performance tensor computations directly on the user's graphics hardware. This approach allows for the execution of large language models locally, ensuring that all data processing remains private to the client d

ChatGLM-6B is an open-source bilingual large language model designed for natural dialogue and text generation in both English and Chinese. It is structured as a dialogue model capable of tasks such as role-playing and information extraction. The project provides implementations for quantized language models, using low-precision weights to reduce GPU memory requirements for local inference. It also supports parameter-efficient fine-tuning, allowing model behavior to be optimized for specific tasks without requiring full retraining. The model includes capabilities for local execution on GPUs a

This repository serves as an educational resource for learning the foundational architectures of natural language processing through concise code implementations. It provides a structured collection of deep learning models designed to process and understand human language, focusing on the core mechanics of neural network sequence modeling and text analysis. The project distinguishes itself by offering direct, hands-on implementations of complex architectures, including Transformers, attention mechanisms, and word embedding generation. By utilizing tensor-based computational graphs and gradien

minGPT is a minimal implementation of the Transformer architecture designed for training and experimenting with language models. It functions as a neural network training framework and a text generation engine, providing the necessary tools to manage data loading, backpropagation, and parameter updates for custom deep learning models. The project is structured as an educational resource for understanding how transformer architectures function by building and training models from scratch. It utilizes a modular block architecture and transformer-based self-attention to process sequences, allowi

This project is a manual reconstruction of the Llama 3 transformer architecture implemented as a PyTorch neural network. It serves as a reference for the internal mathematical structure and tensor flow of a transformer-based language model designed for next token prediction. The implementation focuses on building the model from scratch using basic matrix operations and tensor manipulations. It demonstrates the manual construction of core components, including rotary positional embeddings, multi-head self-attention, and root mean square normalization. The codebase covers the full inference pi

This project is a collection of educational resources and technical guides focused on the development and implementation of large language models. It provides a comprehensive curriculum covering transformer architectures, training methods, and deployment strategies. The materials provide detailed instructions for building autonomous agents using reasoning loops and tool integration, as well as guides for fine-tuning models through supervised learning and preference optimization. It also includes tutorials for constructing retrieval augmented generation pipelines and implementing transformer m

This project is a PyTorch-based framework and implementation suite for the supervised classification of Chinese text. It serves as a deep learning text classifier designed to automate the process of labeling and organizing Chinese language documents into predefined categories. The framework provides a collection of neural network architectures, including TextCNN, Transformer, and FastText. It allows for the selection and prototyping of different model topologies through a modular implementation, enabling the evaluation of various sequence models on specific datasets. The system covers a full

Efficient-AI-Backbones is a lightweight neural network library and computer vision model zoo. It provides a collection of optimized deep learning backbones designed to minimize computational overhead and memory usage for artificial intelligence tasks. The project implements specialized architectures such as GhostNet and MLP to reduce processing requirements. It features a modular backbone design and the distribution of pretrained weights to accelerate the development and deployment of vision models. The library covers efficient neural network design and edge device AI optimization. Its capab

This project is an educational course and set of instructional materials for building large language models from scratch using Python. It provides a step-by-step guide and practical tutorials focused on the internal mechanics of transformer architectures and pre-training workflows. The repository features a framework for implementing and comparing diverse model families, including Llama, GLM, and RWKV. It uses a configuration-driven assembly approach to analyze the structural differences and internal mechanisms of these various architectures. The codebase covers the complete development pipe

Open-Sora-Plan is a text-to-video framework and distributed video training system. It utilizes a diffusion transformer architecture and large language model components to transform written descriptions or image prompts into high-quality video sequences. The system features a distributed infrastructure designed for large-scale video training and inference. It employs sequence parallelism to split high-resolution or long-duration video samples across multiple GPUs and uses a sparse attention mechanism to increase processing speed. The project includes capabilities for both text-to-video and im

Pythia is a multimodal research framework and distributed training system designed for building, training, and evaluating large models that combine visual and linguistic data. It provides a modular environment for developing vision-language models, focusing on the integration of image and text inputs into shared feature representations. The framework utilizes a modular architecture that decouples model building blocks into interchangeable components, allowing for flexible configuration of vision and language modules. It includes a benchmark suite for executing reference models against standar

Lingua is a research framework for developing, training, and experimenting with large language model architectures and data strategies. It provides a lean codebase designed to facilitate the iteration of new model designs through a combination of distributed training orchestration and evaluation pipelines. The framework includes a distributed training orchestrator that generates submission scripts and manages configurations for launching tasks across compute clusters. It utilizes a configuration management system that allows model parameters to be overridden via data classes and command-line

RWKV-LM is a framework for training and deploying recurrent language models. It utilizes a linear-time recurrent architecture that enables text generation and sequence processing with constant memory and time complexity, avoiding the quadratic scaling of traditional attention caches. The project implements a parallelizable training mechanism that allows recurrent models to be trained using global operations while maintaining cache-free inference. It includes state-tuning capabilities to optimize the initial hidden state and utilizes adaptive probability-mass sampling to control token diversit

Stanza is a Python natural language processing library designed for tokenization, lemmatization, and dependency parsing across many human languages using neural models. It provides a neural processing pipeline that converts raw text into structured linguistic data objects, alongside a specialized analyzer for extracting medical insights from clinical and biomedical language. The project includes a wrapper that connects Python scripts to Java-based natural language processing tools and remote annotation servers. This enables a bridge for extracting linguistic annotations and analysis data from

CV-Backbones is a computer vision backbone library and model zoo providing a collection of pre-defined neural network architectures for extracting visual features and processing image data. It serves as a PyTorch vision framework of reusable deep learning components designed for image analysis and visual representation learning. The library focuses on efficient neural network architectures to reduce computational overhead while maintaining feature extraction performance. This is achieved through the implementation of lightweight model designs such as GhostNet and MLP. The project covers a br

Leaf is a machine learning framework and neural network architecture toolkit used for building, training, and deploying models. It functions as a hardware abstraction layer, mapping high-level computational graphs to low-level instructions across various CPU and GPU backends and operating systems. The system enables the design of flexible model structures through a modular architecture where reusable container layers encapsulate weights and mathematical operations. This allows for the composition of complex neural networks via nested components. The framework includes a data engineering pipe

This project is an educational course and learning curriculum for implementing and fine-tuning transformer models using the Hugging Face ecosystem. It serves as a structured guide and technical walkthrough for processing multimodal data, adapting pre-trained neural networks, and deploying models. The material includes a guide for managing, versioning, and distributing model weights and datasets through a centralized asset hub. It also provides a practical tutorial on adapting models to specific datasets using parameter-efficient methods and an implementation guide for solving natural language

ArrayFire is a hardware-agnostic compute framework and JIT-compiled tensor engine designed for high-performance numerical computing. It serves as a GPU numerical computing library and parallel signal processing toolkit that abstracts hardware backends, allowing the same codebase to execute across various GPU architectures and CPUs. The project distinguishes itself through a JIT engine that uses expression compilation to fuse operations and minimize memory overhead. It employs a deferred execution graph to optimize computation chains and provides interoperability primitives to share data and e

ML for Hackers is a machine learning educational resource and library designed for learning the fundamentals of algorithmic programming and data analysis. It provides a neural network framework and a collection of mathematical implementations for building and training predictive models. The project utilizes a modular architecture for stacking linear transformations and activation layers. It implements core deep learning components from scratch using multi-dimensional arrays for tensor algebra and operations. The framework covers a variety of algorithmic capabilities, including automatic diff

TNN is a deep learning inference framework designed to execute pre-trained neural networks across mobile, desktop, and server hardware. It functions as a hardware-accelerated runtime and model compression toolkit, providing a unified interface for deploying models in diverse environments. The framework includes an ONNX model converter to transform models from various training frameworks into a standardized internal format. It distinguishes itself through a combination of model compression tools—including weight quantization and static-code pruning—and a memory management system that reuses bu