# Model Training

Genie Studio is a one-stop development platform specifically designed by Zhiyuan Robotics for embodied AI scenarios, with full-chain product capabilities covering data acquisition, model training, simulation evaluation, and model inference. It provides developers with a standardized solution from 'acquisition' to 'training' to 'testing' to 'inference', greatly reducing the development threshold and improving development efficiency. The platform promotes the rapid development and application of embodied AI technology through efficient data acquisition, flexible model training, precise simulation evaluation, and seamless model inference. Genie Studio not only provides powerful tools but also supports the large-scale implementation of embodied AI, accelerating the industry's leap to a new stage of standardization, platformization, and mass production.

EaseVoice Trainer is a backend project designed to simplify and enhance the speech synthesis and conversion training process. This project is an improvement based on GPT-SoVITS, focusing on user experience and system maintainability. Its design philosophy differs from the original project, aiming to provide a more modular and customizable solution suitable for various scenarios, from small-scale experiments to large-scale production. This tool can help developers and researchers conduct speech synthesis and conversion research and development more efficiently.

The LLMs.txt generator is an online tool powered by Firecrawl, designed to help users generate integrated text files for LLM training and inference from websites. By integrating web content, it provides high-quality text data for training large language models, thereby improving model performance and accuracy. The main advantages of this tool are its simple operation and high efficiency, allowing for the quick generation of required text files. It is primarily aimed at developers and researchers who need a large amount of text data for model training, providing them with a convenient solution.

MLGym is an open-source framework and benchmark developed by Meta's GenAI team and the UCSB NLP team for training and evaluating AI research agents. By offering diverse AI research tasks, it fosters the development of reinforcement learning algorithms and helps researchers train and evaluate models in real-world research scenarios. The framework supports various tasks, including computer vision, natural language processing, and reinforcement learning, aiming to provide a standardized testing platform for AI research.

kg-gen is an artificial intelligence-based tool that extracts knowledge graphs from plain text. It supports text inputs ranging from single sentences to lengthy documents and can handle conversational-style messages. Leveraging advanced language models and structured output techniques, this tool helps users quickly construct knowledge graphs, suitable for natural language processing, knowledge management, and model training, among other applications. kg-gen provides flexible interfaces and a variety of functionalities designed to streamline the knowledge graph generation process and enhance efficiency.

Steev is a tool specifically designed for AI model training, aimed at simplifying the training process and enhancing model performance. It automatically optimizes training parameters, provides real-time monitoring of the training process, and offers code reviews and suggestions to help users complete model training more efficiently. One of Steev's main advantages is its ease of use, requiring no configuration, making it suitable for engineers and researchers aiming to improve model training efficiency and quality. Currently in a free trial phase, users can experience all its features at no cost.

Kolosal AI is a tool for training and running large language models (LLMs) on local devices. By streamlining the processes of model training, optimization, and deployment, it enables users to leverage AI technology efficiently on local hardware. The tool supports various hardware platforms, provides fast inference speeds, and offers flexible customization capabilities, making it suitable for a wide range of applications from individual developers to large enterprises. Its open-source nature also allows users to conduct secondary development according to their specific needs.

Open Thoughts is a project led by Bespoke Labs and the DataComp community, aimed at curating high-quality open-source reasoning datasets for training advanced small models. The project brings together researchers and engineers from various universities and research institutions, including Stanford University, the University of California, Berkeley, and the University of Washington, dedicated to promoting the development of reasoning models through high-quality datasets. The project is established in response to the growing demand for applications of reasoning models in fields such as mathematics and code reasoning, where high-quality datasets are critical for improving model performance. Currently free to access, the project primarily targets researchers, developers, and professionals interested in reasoning models, with its open-source datasets and tools serving as a significant resource for advancing AI education and research.

Flock of Finches 37B-A11B v0.1 is the latest member of the RWKV family, representing an experimental model with 1.1 billion active parameters. Despite being trained on only 109 billion tokens, it performs comparably to the recently released Finch 14B model on common benchmark tests. This model employs an efficient sparse mixture of experts (MoE) approach, activating only a portion of parameters for any given token, thereby saving time and reducing computational resource usage during training and inference. Although this architectural choice incurs higher VRAM usage, from our perspective, it is highly beneficial to train and operate a model with greater capacity at a lower cost.

E2M is a Python library capable of parsing and converting multiple file types into Markdown format. It employs a parser-converter architecture, supporting the conversion of a variety of file formats including doc, docx, epub, html, htm, url, pdf, ppt, pptx, mp3, and m4a. The ultimate aim of the E2M project is to provide high-quality data for Retrieval-Augmented Generation (RAG) and model training or fine-tuning.

TRELLIS is a native 3D generation model based on a unified structured latent representation and a correction transformer, capable of producing diverse and high-quality 3D assets. The model captures structural (geometric) and texture (appearance) information comprehensively by integrating sparse 3D meshes with dense multi-view visual features extracted from powerful visual foundation models, while maintaining flexibility during the decoding process. TRELLIS can handle up to 2 billion parameters and has been trained on a large dataset of 3D assets containing 500,000 diverse objects. It generates high-quality results conditioned on text or images, significantly surpassing existing methods, including recent approaches of similar scale. TRELLIS also demonstrates flexible output format options and local 3D editing capabilities, which were not provided by previous models. Source code, models, and data will be made available.

PrimeIntellect-ai/prime is a framework designed for efficient, globally distributed training of AI models over the internet. Through technological innovation, it facilitates cross-regional AI model training, improves computing resource utilization, and reduces training costs, which is critical for AI research and application development that requires significant computational resources.

MM1.5 is a series of multimodal large language models (MLLMs) designed to enhance capabilities in understanding text-rich images, visual reference grounding, and multi-image reasoning. Based on the MM1 architecture, the model adopts a data-centric training approach and systematically explores the impact of different data mixes throughout the model training lifecycle. The MM1.5 model varies from 1B to 30B parameters and includes both dense and mixture of experts (MoE) variants, providing valuable guidance for future MLLM development research through extensive empirical and ablation studies that detail the training processes and decision insights.

RECE is a concept erasure technology for text-to-image diffusion models that reliably and efficiently removes specific concepts by introducing regularization terms during model training. This technology is significant in enhancing the safety and controllability of image generation models, especially in scenarios where generating inappropriate content needs to be avoided. The primary advantages of RECE technology include high efficiency, high reliability, and easy integration into existing models.

Flux Gym is a streamlined Web UI designed for FLUX LoRA model training, particularly suited for devices with 12GB, 16GB, or 20GB VRAM. It combines the user-friendliness of the AI-Toolkit project with the flexibility of Kohya Scripts, enabling users to train models without complex terminal operations. Flux Gym allows users to upload images and add descriptions through a simple interface, then start the training process.

Easy Voice Toolkit is an AI voice toolkit based on open-source voice projects, providing various automated audio tools including speech model training. The toolkit seamlessly integrates to create a complete workflow, allowing users to selectively use these tools or utilize them in sequence to gradually convert raw audio files into ideal speech models.

Civita Green is a community platform for AI enthusiasts, artists, and developers, offering AI model training, image and video creation, and art sharing. The platform supports users in creating, sharing, and utilizing various AI models to foster the development of AI art.

ai-toolkit is a research-oriented GitHub repository created by Ostris, primarily used for experiments and training with Stable Diffusion models. It contains a variety of AI scripts that support model training, image generation, LoRA extraction, and more. The toolkit is still in development and may have some instability, but it offers rich features and high customizability.

x-flux is a collection of deep learning model training scripts released by the XLabs AI team, featuring LoRA and ControlNet models. These models leverage DeepSpeed for training and support image sizes of 512x512 and 1024x1024, along with corresponding training configuration files and examples. The x-flux model training aims to enhance the quality and efficiency of image generation, making it significant in the field of AI image generation.

This GitHub repository contains training and inference code to replicate our winning solution in the AI Mathematics Olympic (AIMO) Progress Prize 1. Our solution consists of four main parts: a recipe for fine-tuning DeepSeekMath-Base 7B for use in solving math problems using Tool Integrated Reasoning (TIR); two high-quality datasets of about 10 million math questions and solutions; an algorithm for generating solution candidates with coding execution feedback (SC-TIR); and four carefully selected validation sets from AMC, AIME, and MATH to guide model selection and avoid overfitting the public leaderboard.

Prime Intellect is committed to democratizing AI development on a scalable scale. It offers the discovery of global computing resources, model training, and the capability to co-own smart innovation. By distributing training across clusters, it enables users to train cutting-edge models and co-own the open AI innovation outcomes, including language models and scientific breakthroughs.

Prov-GigaPath is a whole-slide foundation model for digital pathology research. Trained on real-world data, it aims to support AI researchers in their studies of pathology foundational models and digital pathology slide data encoding. Developed by multiple authors and published in Nature, it is not suitable for clinical care or any clinical decision-making purposes and is restricted to research use only.

CoreNet is a deep neural network toolkit that enables researchers and engineers to train both standard and innovative small to large-scale models for a variety of tasks, including foundational models (such as CLIP and LLM), object classification, object detection, and semantic segmentation.

LlamaParse, part of the LLAMA project, is a tool for parsing and processing related data. LLAMA is a library for machine learning models, focusing on providing user-friendly interfaces and efficient data processing capabilities.

DataDreamer is a powerful open-source Python library for prompt engineering, synthetic data generation, and training workflows. Designed for simplicity, extreme efficiency, and research-grade quality, DataDreamer supports creating prompt workflows, generating synthetic datasets, aligning and fine-tuning models, instruction tuning, model distillation, and simplifies the sharing and reproducibility of datasets and models.

Meta has released the Video Joint Embedding Predictive Architecture (V-JEPA) model, a significant step forward in advancing machine intelligence, providing a more pragmatic understanding of the world.

StemGen is an end-to-end music generation model trained to listen to musical contexts and generate appropriate responses. It is built upon a non-autoregressive language modeling architecture, similar to SoundStorm and VampNet. More details can be found in the paper. This page showcases various example outputs of this architecture model.

AXLearn is a deep learning library built by Apple based on JAX and XLA. It takes an object-oriented approach to address software engineering challenges in large-scale deep learning model development. Its configuration system allows users to combine models from reusable building blocks and integrate with other libraries (such as Flax and Hugging Face transformers). AXLearn aims to scale training, supporting models with hundreds of billions of parameters efficiently trained on thousands of accelerators, suitable for deployment on public clouds. It also adopts a global computation paradigm, allowing users to describe computations on a global virtual machine instead of individual accelerators. AXLearn supports a wide range of applications, including natural language processing, computer vision, and speech recognition, and includes baseline configurations for training state-of-the-art models.

Datasaur is a leading NLP data annotation platform that can accelerate project speeds by 10x and improve model performance by 2x. It offers configurable annotation, advanced quality control, and automation features, empowering engineers to focus on building high-quality models.

Volcano Ark provides comprehensive functions and services for model training, inference, evaluation, and fine-tuning, and focuses on supporting the large model ecosystem. Curated models ensure model stability, a rich platform of applications and tools, information security, powerful computing capabilities, and professional services. Key functions include Model Marketplace, Model Experience, Model Training & Inference, and Model Applications. Suitable for application scenarios in industries such as automobiles, finance, consumer goods, the broad internet, and education & office.