Projects sharing features with Video Pre Training

Lab is a customizable 3D platform and research testbed designed for training and testing autonomous agents using reinforcement learning. It serves as a spatial AI training simulator where agents can be evaluated through navigation and puzzle-solving tasks. The environment allows for the definition of complex layouts and task behaviors through external scripting, enabling the generation of specific challenges for AI research. It supports both automated training via standard API bindings and manual agent control to validate simulation dynamics. The system utilizes a grid-based spatial represen

giga-brain-0 is a robot action model framework designed to train and deploy neural networks that map multi-modal sensor data to physical robot control signals. It functions as a robot manipulation controller that processes high-dimensional observations to execute dexterous, long-horizon physical tasks. The project provides a multi-modal robot inference server using a client-server architecture to stream real-time vision and language observations for instant action prediction. It includes an embodiment fine-tuning pipeline to adapt pre-trained base models to specific robot hardware configurati

This project is a PyTorch implementation of 3D residual networks designed for video action recognition. It provides a spatiotemporal architecture that analyzes both spatial frames and temporal motion to classify human activities within video clips. The system includes a distributed model training framework to accelerate learning across multiple compute nodes. It supports the deployment and fine-tuning of pre-trained model weights, allowing the adaptation of existing networks to specific new datasets. The codebase covers the full pipeline for spatiotemporal learning, including video dataset p

This project is an educational course and collection of training materials focused on generative diffusion models. It provides a curriculum and practical guides for training, fine-tuning, and deploying models capable of synthesizing images, audio, and video. The material covers specific implementation strategies including noise-based synthesis, iterative refinement, and latent space compression. It provides instruction on guiding generative outputs through conditional synthesis and prompt adherence optimization, as well as techniques for image inpainting and text-based editing. The project i

Stable-audio-tools is a toolkit for training and deploying latent diffusion models for high-fidelity audio synthesis. It provides a framework for generating audio by iteratively refining noise within a compressed latent space, using specialized encoders to preserve temporal and spectral features of the audio signal. The project features a system for adapting pre-trained audio checkpoints to new datasets through modular initialization and configuration files. It includes utilities for weight extraction and inference model export, which remove training metadata and optimizer states to create li

OpenManus-RL is a reinforcement learning framework and distributed training pipeline designed to train large language models as agents. It serves as an agentic reasoning optimizer and reward model trainer, providing the infrastructure to improve model decision-making through reward-based policy optimization. The project distinguishes itself through a distributed architecture that supports parameter sharding across multiple compute nodes and a coordinated rollout system for collecting interaction trajectories. It incorporates advanced reasoning strategies, such as Tree-of-Thoughts and Monte Ca

CleanRL is a reinforcement learning library and PyTorch framework providing a suite of reproducible implementations for online reinforcement learning algorithms. It serves as a deep reinforcement learning benchmark suite and experiment orchestrator designed for research and agent development across both discrete and continuous action spaces. The project is distinguished by its single-file algorithm implementation approach, which encapsulates each algorithm in a standalone script to eliminate complex class hierarchies. This structure is paired with a system for scheduling and executing large-s

DeepMimic is a deep reinforcement learning framework and physics-based motion imitation tool designed to teach simulated characters and robots to reproduce human movements. It provides a pipeline for integrating motion capture data into physics simulations to train agents that can mimic complex physical skills. The system utilizes the PyBullet simulation environment to execute motion policies and visualize character interactions in real time. It includes a motion capture integration pipeline that imports and processes animation sequences to serve as reference targets for imitation learning ag

Dopamine is a reinforcement learning research framework designed for prototyping and testing algorithms across diverse simulated environments. It provides an agent development toolkit that utilizes a flat class hierarchy to facilitate the creation and extension of learning agents. The framework includes a standardization layer via environment wrappers that connect agents to various physics simulations and gaming environments. It also features a high-performance experience replay buffer for storing and sampling transition data to improve training stability, alongside a dedicated hyperparameter

ManiSkill is a GPU-accelerated robot simulation framework designed for training robotic manipulation skills, benchmarking learning algorithms, and generating synthetic datasets. It serves as a reinforcement learning environment where robot control policies can be developed and evaluated using parallelized physics and rendering on the GPU. The platform is distinguished by its ability to perform sim-to-real transfer, allowing policies trained in virtual environments to be deployed onto physical robotic hardware. It features ray-traced parallel rendering for producing high-frame-rate RGBD and se

SO-ARM100 is an open-source robot arm hardware project providing 3D-printable designs and assembly guides for building affordable robotic arms. It includes calibration software to synchronize motor communication parameters and arm positions via USB, alongside hardware designs for tactile sensing robotic grippers. The project distinguishes itself through the integration of touch-sensing and flexible filaments for adaptive grasping. It also provides a dedicated imitation learning dataset tool, featuring a web interface for labeling and visualizing robotics data to train machine learning models

Tensorlayer is a deep learning framework and cross-backend AI library used to construct and execute neural network models. It serves as a scientific neural network toolkit providing customizable layers and architectures designed for research applications in science and engineering. The library enables multi-backend model execution, allowing the same model code to run across different deep learning frameworks, GPUs, and specialized AI accelerators. It includes a reinforcement learning library that provides both low-level and high-level tools for developing intelligent agents.

Open Spiel is a research library and framework for reinforcement learning, planning, and multi-agent game simulation. It provides a system for representing single-agent and multi-agent games across zero-sum, cooperative, and imperfect information scenarios. The project utilizes a standardized abstract game interface to decouple game logic from algorithms, allowing agents to run across different game types. It implements performance-critical logic in C++ with Python bindings and uses deterministic seeded simulation to ensure reproducibility for research benchmarking. The framework covers a br

TradeMaster is a reinforcement learning trading framework and algorithmic trading simulator designed for designing and testing quantitative trading strategies. The system provides a platform for developing reinforcement learning agents, managing quantitative portfolios, and optimizing trade execution using financial market data. The project features specialized components for multi-modality data preprocessing, a high-fidelity market environment simulation for strategy backtesting, and a quantitative portfolio manager for capital reallocation across multiple assets. It includes a trade executi

This project is a reinforcement learning framework and game AI engine designed for training adversarial agents in two-player turn-based games. It implements a training loop that utilizes self-play and Monte Carlo Tree Search to produce neural networks capable of predicting board strength and move probabilities. The system decouples the reinforcement learning engine from specific game rules through an abstract game logic interface, allowing for the definition of custom game rules, win conditions, and board representations. It supports integration with various deep learning frameworks to serve

PufferLib is a reinforcement learning framework built around high-speed environment simulation and automatic hyperparameter optimization. It is designed to accelerate the entire RL training pipeline by running simulations at near-native speed and enabling the training of tiny models to super-human performance within seconds. The framework achieves its speed through a single-process training loop that eliminates inter-process communication overhead, vectorized batched simulation for parallel environment execution, and compiled C extensions that offload performance-critical computations. It als

The Arcade Learning Environment is a research-focused platform that provides a high-performance emulation engine for Atari 2600 games. It serves as a standardized environment for training and evaluating artificial intelligence agents, bridging native game execution with machine learning libraries through a dedicated Python interface. The project distinguishes itself by offering robust support for parallel and vectorized environment execution, allowing multiple game instances to run simultaneously to accelerate data collection. It includes specialized tools for asynchronous command queuing and

ART is a platform for agentic training, providing a reinforcement learning framework, training environment, and compute orchestrator. It enables the improvement of multi-step agent reasoning and tool usage through group relative policy optimization and a judge-based reward modeling system. The project features tools for model distillation to transfer capabilities from large teacher models to smaller architectures, as well as a system for capturing execution trajectories to generate synthetic training data. It supports specialized training workflows including supervised fine-tuning for baselin

Oasis is an LLM-powered multi-agent social simulator and research tool designed to study synthetic social phenomena. It functions as a synthetic social network platform, replicating the infrastructure of social sites including user profiles, follow relationships, and content discovery mechanisms to model human-like social behaviors at scale. The framework orchestrates large-scale agent populations, supporting up to one million autonomous agents. It distinguishes itself by translating language model outputs into concrete social actions and external tool executions through a tool-calling orches

AI Town is a TypeScript-based simulation engine used to create virtual environments where autonomous characters interact and socialize. It functions as a framework for orchestrating multiple AI agents within a persistent digital world, utilizing language models and a game engine to drive character behavior and social interactions. The project differentiates itself through a dedicated agent sandbox and a vector database agent store, which allow for the management of agent memories and world state. It integrates generative AI for background music and provides tools for simulation world design,

Legged Gym is a high-performance simulation platform and toolkit engineered for training autonomous robotic agents in complex, physics-based environments. It provides a comprehensive framework for developing legged locomotion control policies, enabling robots to learn movement strategies for navigating uneven terrain and managing physical disturbances through reinforcement learning. The platform distinguishes itself by utilizing hardware-accelerated physics and headless execution to maximize computational throughput during training. It incorporates a domain randomization pipeline that injects

MedSAM is a deep learning framework designed for automating the segmentation of anatomical structures in 2D and 3D medical imagery. It provides specialized tools for fine-tuning pretrained segmentation weights on custom medical datasets and evaluating the accuracy of those predictions against ground truth labels. The project focuses on adapting the Segment Anything Model architecture for medical use, enabling the isolation of specific anatomical structures through prompt-guided methods such as bounding boxes and point prompts. The system covers a full medical AI workflow, including data engi

This project is an AI research implementation library and machine learning research repository. It provides a collection of reference code, illustrative implementations, and open-source research datasets used to verify hypotheses and build upon existing models in artificial intelligence. The repository focuses on scientific research reproduction by translating theoretical findings from published papers into executable code. It includes specialized scientific simulation environments designed to test the behavior of autonomous agents and models within controlled settings. The project covers AI

bert4keras is a lightweight reimplementation of the BERT transformer architecture for the Keras deep learning framework. It serves as a natural language processing toolkit and transformer model library used for text classification, sequence labeling, and semantic embedding extraction. The framework includes a sequence-to-sequence model system for question answering and text generation, as well as a model inference server to deploy trained transformers as web APIs for real-time predictions. Capabilities cover a broad range of natural language understanding tasks, including reading comprehensi

ParlAI is a conversational AI research framework designed for training, evaluating, and sharing dialogue models using a unified interface for datasets and agents. It functions as a PyTorch-based training platform and a dialogue data collection system, providing a centralized model zoo for the distribution of versioned pretrained agents. The project distinguishes itself through a knowledge-grounded retrieval system that combines dense and sparse indexing to ground responses in external information. It also provides a comprehensive infrastructure for gathering human-AI interaction data via inte

This repository is a collection of frameworks and guides for Llama models, functioning as a fine-tuning framework, an inference pipeline, and an AI workflow orchestrator. It provides tools for adapting large language models to specific datasets and domains. The project includes a parameter-efficient fine-tuning toolkit that utilizes techniques like low-rank adaptation to reduce memory and compute requirements. It also serves as an implementation guide for retrieval-augmented generation, combining model inference with external data retrieval to improve response accuracy. The capability surfac