Projects

Overview: Cambrian-1 is a family of multimodal language models designed with a focus on vision-centric features. This project aims to explore and evaluate various aspects of multimodal language models, particularly in the context of vision-centric design. Key features: - Multi-modal language models with a vision-centric design. - Support for evaluation benchmarks. - Targeted data engine for efficient data processing. - Release of Cambrian-1 models (8B, 13B, 34B) and training data. Tech stack: Python Why it matters: This project provides a comprehensive framework for evaluating and exploring the capabilities of multimodal language models, with a specific emphasis on vision-centric design, making it valuable for researchers and developers interested in advancing this field. Repository: cambrian-mllm/cambrian GitHub stars: 2011 GitHub description: Cambrian-1 is a family of multimodal LLMs with a vision-centric design. README excerpt: Cambrian-1 is a family of multimodal language models designed with a focus on vision-centric features. This project aims to explore and evaluate various aspects of multimodal language models, particularly in the context of vision-centric design.

### Overview: This project introduces the Keras Temporal Convolutional Network (TCN), a specialized architecture designed for processing sequences of data. The TCN is particularly effective for tasks involving long-term dependencies and can outperform traditional recurrent neural networks (RNNs) in certain scenarios. ### Key Features: - **Supports Python and R**: The TCN is compatible with both Python and R, making it versatile for various applications. - **Fully Working Example**: A fully working example is provided using R, demonstrating how to integrate the TCN into a real-world scenario. - **Performance Comparison**: The TCN is shown to perform better than LSTM/GRU on tasks such as sequence MNIST, adding problems, copy memory, and word-level PTB, highlighting its advantages in handling long sequences. ### Tech Stack: The project utilizes Keras, a high-level neural networks API, and TensorFlow, a popular open-source machine learning framework. ### Why It Matters: The TCN architecture offers several advantages over traditional RNNs, including better memory capacity, improved performance on long sequences, and enhanced stability in gradients, making it a valuable tool for sequence processing tasks in both research and industry applications.

Overview: IELTS Atlas is a pure-frontend practice system designed for IELTS reading, offering question-bank browsing, reading practice, optional listening practice, test-set practice, practice records, score statistics, mistake analysis, data backup, question-bank import, vocabulary tools, reading review, and more. Key features: - Question-bank browsing - Reading practice - Optional listening practice - Test-set practice - Practice records - Score statistics - Mistake analysis - Data backup - Question-bank import - Vocabulary tools - Reading review Tech stack: Pure frontend using HTML, CSS, and JavaScript bundles. Why it matters: IELTS Atlas provides a comprehensive and user-friendly practice system for IELTS reading, making it an essential tool for students aiming to improve their IELTS scores.

Overview: DIG, Dive into Graphs, is a comprehensive library designed for graph deep learning research. It offers unified implementations for various advanced tasks such as graph generation, self-supervised learning, explainability, 3D graph processing, and more. Key features: - Unified data interfaces - Common algorithms - Evaluation metrics Tech stack: Python Why it matters: DIG simplifies the process of developing and benchmarking graph deep learning methods, making it easier for researchers to experiment with new ideas and compare results with established benchmarks. Repository: divelab/DIG Primary language: Python GitHub stars: 2009 GitHub description: A library for graph deep learning research Key features: - Unified data interfaces - Common algorithms - Evaluation metrics Tech stack: Python Why it matters: DIG simplifies the process of developing and benchmarking graph deep learning methods, making it easier for researchers to experiment with new ideas and compare results with established benchmarks.

### Overview: LangAlpha is an open-source AI agent designed for financial market research and trading. It leverages Claude Code to research the market, build a thesis, and trade your own brokerage account within specified permissions. ### Key Features: - **Research with a Team of Agents:** LangAlpha splits research questions across multiple analysts, leveraging Cerebras (CBRS) to pull filings, prices, options, and macro data, and run the numbers in code. - **Thesis Workspace:** The research results are kept in a workspace where analysts can build on the work, with the ability to send follow-up questions to redirect them mid-run. - **Thesis Thesis:** The final thesis is built from the research and is kept in a workspace where analysts can build on the work, with the ability to send follow-up questions to redirect them mid-run. - **Thesis Watch:** The final thesis is built from the research and is kept in a workspace where analysts can build on the work, with the ability to send follow-up questions to redirect them mid-run. - **Thesis Watch:** The final thesis is built from the research and is kept in a workspace where analysts can build on the work, with the ability to send follow-up questions to redirect them mid-run. - **Thesis Watch:** The final thesis is built from the research and is kept in a workspace where analysts can build on the work, with the ability to send follow-up questions to redirect them mid-run. - **Thesis Watch:** The final thesis is built from the research and is kept in a workspace where analysts can build

### Overview: MMGeneration is a comprehensive toolkit for generative models, designed to be used with PyTorch and MMCV. It supports a wide range of tasks and models, including text-to-image generation, 3D-aware generation, and more. ### Key Features: - **Text2Image:** Supports GLIDE, Disco-Diffusion, and Stable-Diffusion models. - **3D-aware Generation:** EG3D model. - **GAN Interpolation, Projection, and Manipulations:** Integrated into the framework for easy manipulation of GANs. ### Tech Stack: Python ### Why It Matters: MMGeneration provides a powerful and flexible toolkit for generative models, making it an essential tool for researchers and developers in the field of artificial intelligence and computer vision. Its support for various models and tasks makes it a valuable resource for anyone looking to explore and advance generative models.

### Overview: FireRedASR is an open-source industrial-grade Automatic Speech Recognition (ASR) model designed for Mandarin, Chinese dialects, and English. It achieves state-of-the-art performance on public Mandarin ASR benchmarks and excels in singing lyrics recognition. ### Key Features: - **ASR, VAD, LID, and Punc Modules**: All modules perform at the state-of-the-art (SOTA) level. - **Seamless End-to-End Speech Interaction**: Designed for high performance and efficiency. - **State-of-the-Art (SOTA) Performance**: Achieves outstanding results on public Mandarin ASR benchmarks. - **Singing Lyrics Recognition**: Exceptional capability for recognizing lyrics in singing voices. ### Tech Stack: Python ### Why It Matters: FireRedASR offers unparalleled performance and efficiency across various applications, making it a valuable tool for developers and researchers in the field of ASR. Its ability to handle diverse languages and dialects sets it apart, making it a key asset for projects requiring high accuracy and reliability in ASR tasks. Repository: FireRedTeam/FireRedASR GitHub Stars: 2006 GitHub Description: Open-source industrial-grade ASR models supporting Mandarin, Chinese dialects, and English, achieving a new SOTA on public Mandarin ASR benchmarks, while also offering outstanding singing lyrics recognition capability.

### Overview: Ante is a self-contained agent harness designed to work with any model, offering a lightweight and efficient solution for coding tasks. It is optimized for performance and minimal resource usage, making it ideal for developers and coders looking for a fast and reliable tool. ### Key Features: - **Self-contained Agent Harness**: Ante is a single executable that can run on any operating system, including macOS, Linux, and Windows. - **Zero Runtime Dependencies**: No external libraries or dependencies are required, ensuring a clean and lightweight installation. - **Dynamic Model Integration**: Ante can seamlessly integrate with various models, allowing for a flexible and adaptable solution. - **Cross-Model Comparison**: Ante is evaluated across different model families, providing insights into its performance and adaptability. ### Tech Stack: Rust ### Why It Matters: Ante is designed to be a versatile and efficient solution for coding tasks, offering a lightweight and resource-efficient approach. Its dynamic model integration capabilities make it suitable for a wide range of applications, and its continuous evaluation ensures it remains at the forefront of performance and adaptability.

### Overview: Patterns AI Core's LangchainRB project is a Ruby-based framework for building language model-powered applications. It simplifies the integration of various large language models into Ruby-based Rails applications, making it easier to develop applications that leverage advanced natural language processing capabilities. ### Key Features: - **Unified Interface**: Provides a consistent API for interacting with different large language model providers, allowing developers to easily switch between providers without modifying their application code. - **LLM Providers**: Supports a wide range of LLM providers including Anthropic, AWS Bedrock, Azure OpenAI, Cohere, Google Gemini, Google Vertex AI, HuggingFace, Mistral AI, Ollama, OpenAI, Replicate, and more. - **Prompt Management**: Offers tools for managing and formatting prompts, making it easier to craft effective input for LLMs. - **Output Parsers**: Provides parsers for different output formats, ensuring that the generated responses can be easily integrated into the application. ### Tech Stack: Ruby ### Why it Matters: This project is essential for developers looking to integrate advanced natural language processing into their Ruby-based Rails applications. It simplifies the integration process, making it easier to build applications that can understand and respond to human language in a more sophisticated way.

Overview: Tokencost is a Python library designed to estimate the cost of using large language model APIs, specifically focusing on the token cost for prompts and completions. Key features: - Tracks the latest price changes for major LLM providers. - Accurately counts the tokens in prompts and completions. - Provides a simple function to calculate the cost of a prompt or completion. Tech stack: Python Why it matters: For developers and AI enthusiasts working with large language models, Tokencost helps in accurately estimating the cost of using these models, which is crucial for budgeting and resource management in applications and AI agents. Repository: AgentOps-AI/tokencost Primary language: Python GitHub stars: 2004 GitHub description: Easy token price estimates for 400+ LLMs. TokenOps.

### Overview PyGPT is an open-source desktop AI assistant designed for Linux, Windows, and macOS. It leverages models from OpenAI (GPT-6 Astra, GPT-5.6, GPT-4), Google Gemini, Anthropic Claude, xAI Grok, Perplexity / Sonar, DeepSeek, and more. The assistant supports various workflows, including chat, agents, real-time voice, and image/video generation, making it versatile for both personal and professional use. ### Key Features - **Chat Interface**: A conversational chat interface for seamless interaction. - **Agents**: Supports multiple work modes, including Chat, Agents, Realtime + audio, Research, Completion, Image and Video generation, Computer use, Experts, Autonomous mode, and Custom agents. - **Model Support**: Runs on local and remote tools, works with files, calls APIs, and controls the desktop or browser. - **Vector-Store**: Integrated RAG for files, webpages, Google/GitHub data, media, images, conversation history, and formats such as `txt`, `pdf`, `csv`, `html`, `md`, `docx`, `json`, `epub`, `xlsx`, and `xml`. - **Image and Video Generation**: Models like `gpt-image`, `Imagen`, `Gemini`, and `Nano Banana` for image generation. - **Web Search**: Access to `DuckDuckGo`, `Google`, `Microsoft Bing`, and remote web search. - **Speech Synthesis and Recognition**: Uses `Open

### Overview: Elyra is a set of AI-centric extensions to JupyterLab Notebooks. It includes features such as an AI-centric pipeline editor, ability to run notebooks as batch jobs, reusable code snippets, AI assistant integration, hybrid runtime support, Python and R script editors, and more. ### Key Features: - **AI-Pipelines Visual Editor** - **Batch Job Execution** - **Reusable Code Snippets** - **AI Assistant Integration** - **Hybrid Runtime Support** - **Python and R Script Editors** - **Python Script Navigation** - **Experimental Python Script Debugger** - **Notebook Navigation Using Table of Contents** - **Language Server Protocol Integration** - **Version Control Using Git Integration** ### Tech Stack: Python ### Why It Matters: Elyra extends JupyterLab with an AI-centric approach, making it easier to manage and run AI-driven workflows within the familiar Jupyter environment.

### Overview: This project is a PyTorch implementation designed for exploring knowledge distillation (KD) experiments with flexibility. It provides a framework for training and evaluating both shallow and deep knowledge distillation models, including hyperparameter management, progress tracking, and checkpoint saving. ### Key Features: - A unified framework for exploring "shallow" and "deep" knowledge distillation experiments. - Hyperparameters defined in a "params.json" file (avoiding long argparser commands). - Hyperparameter searching and result synthesizing. - Progress bar, tensorboard support, and checkpoint saving/loading utilities. - Pretrained teacher models available for download. ### Tech Stack: PyTorch ### Why It Matters: This project is valuable for researchers and developers interested in exploring knowledge distillation techniques for efficient hardware solutions. It provides a flexible framework for experimenting with different models and hyperparameters, making it suitable for various research and development scenarios.

### Overview: 这是一个完整的 Claude Code 配置集合，包括技能、本能、内存优化、持续学习、安全扫描以及研究优先的开发模式。源自 Anthropic 黑客松获胜者的实战配置，旨在帮助中文工程师高效理解与使用 Claude Code。 ### Key features: - 技能 (Skills) - 本能 (Instincts) - 内存优化 (Memory Optimization) - 持续学习 (Continuous Learning) - 安全扫描 (Security Scan) - 研究优先的开发模式 (Prioritized Research Development Mode) ### Tech stack: JavaScript ### Why it matters: 助力中文工程师高效理解与使用 Claude Code，源自 Anthropic 黑客松获胜作品。适用于 Claude Code, Codex, Cowork 等 AI 智能体框架。

### Overview: MemoraX Code is a memory plugin for AI coding that enhances your workflow by turning your engineering experience, repository knowledge, and working style into useful memory. It ensures that you don’t lose context or start over, making every interaction a starting point for the next. ### Key Features: - **Context Retention**: MemoraX Code remembers your past projects, repository knowledge, and working preferences to ensure continuity. - **Memory Layer**: It integrates with Codex, Claude Code, CodeBuddy CLI, WorkBuddy, DeepSeek Harness, OpenCode, Trae, Cursor, and Qoder, providing a shared memory layer for context. - **Relevant Memory**: It focuses on the small amount of memory relevant to the current task, enabling quick and accurate investigation and validation. ### Tech Stack: - **JavaScript** ### Why It Matters: MemoraX Code streamlines your coding process by ensuring that your past projects, repository knowledge, and working style are remembered and reused, making future tasks more efficient and less time-consuming.

### Overview: DSHA 是一款免 ROOT、免 Termux 的 DeepSeek Harness 安卓启动器，提供完整的 Ubuntu 环境，包括 apt 能用、交互式 PTY 能用、原生模块编译通过。无需 Termux、ROOT 或编译，只需安装 APK 即可使用。 ### Key features: - **免 ROOT、免 Termux**：无需 ROOT 或 Termux 来安装和运行。 - **完整的 Ubuntu 环境**：包含 apt 能用、交互式 PTY 能用、原生模块编译通过。 - **实时 AI 输出**：AI 输出实时显示在屏幕上。 - **ADB 直连**：支持 ADB 直连功能。 - **数据不丢**：数据在设备上不丢失。 ### Tech stack: Java ### Why it matters: DSHA 提供了一个简单、安全、高效的 DeepSeek Harness 安卓启动器，无需 ROOT、Termux 或编译，只需安装 APK 即可使用。它为用户提供了一个完整的 Ubuntu 环境，包括 apt 能用、交互式 PTY 能用、原生模块编译通过，支持 ADB 直连和实时 AI 输出，确保数据不丢失。

### Overview: This repository contains custom pipelines and models designed to enhance spaCy for processing scientific and biomedical documents. It includes a custom tokenizer, POS tagger, syntactic parser, and entity span detection models trained on biomedical data. ### Key Features: - Custom tokenizer - POS tagger - Syntactic parser - Entity span detection model - NER models for specific tasks ### Tech Stack: Python ### Why it Matters: This repository is essential for anyone working with scientific or biomedical text data, providing advanced capabilities to better understand and analyze such content.

### Overview: CVNets is a comprehensive library designed for training computer vision models, including object classification, detection, segmentation, and foundation models like CLIP. ### Key features: - Supports various tasks including object classification, object detection, semantic segmentation, and foundation models. - Includes models like MobileNetv1, MobileNetv2, MobileNetv3, EfficientNet, ResNet, RegNet, and CLIP. - Provides examples for training and evaluating models, as well as converting PyTorch models to CoreML. - Enhances distillation support with Soft and Hard distillation techniques. ### Tech stack: Python ### Why it matters: CVNets simplifies the process of training computer vision models, making it accessible for researchers and engineers without requiring deep expertise in deep learning frameworks. ### Repository: apple-aiml-research/ml-cvnets ### GitHub stars: 1996 ### GitHub description: CVNets: A library for training computer vision networks

### Overview: Titans is an unofficial implementation of the state-of-the-art memory module for transformers, known as Titans, in Pytorch. This project aims to explore and enhance the capabilities of the memory module beyond its simple 1-4 layer MLP configuration. ### Key Features: - **Memory Module Enhancements**: The project includes modifications to the memory module to improve its performance and effectiveness. - **Transformer Integration**: The implementation integrates the memory module with a transformer architecture, allowing for more complex and expressive memory structures. - **Experimental Exploration**: The repository contains experiments and explorations into architectures beyond the simple 1-4 layer MLP configuration, with a focus on improving performance and memory efficiency. ### Tech Stack: Pytorch ### Why it Matters: Titans is a significant contribution to the field of transformer memory, offering improved performance and memory efficiency compared to existing methods. By exploring and enhancing the memory module, this project aims to push the boundaries of transformer models and potentially lead to more advanced applications in natural language processing and beyond. ### GitHub Stars: 1996 ### Install: To use Titans, you can install it via pip: ```bash $ pip install titans-pytorch ``` ### Usage: Here's a basic example of how to use the memory module: ```python import torch from titans_pytorch import NeuralMemory mem = NeuralMemory( dim = 384, chunk_size = 64 # set to smaller chunk size for better perf on smaller sequence lengths (but more memory usage) ).cuda() seq = torch.randn(

### Overview: This repository is a collection of research papers and software focused on explainability in graph machine learning. It aims to provide a comprehensive resource for understanding and implementing explainable graph reasoning techniques. ### Key Features: - Research papers on explainable graph reasoning - Software tools for implementing explainable graph reasoning - A survey of theory and papers related to explainable graph reasoning ### Tech Stack: Python, JavaScript, and various machine learning libraries. ### Why it Matters: This collection is crucial for researchers and practitioners looking to enhance the interpretability of graph machine learning models, ensuring that their decisions are transparent and understandable.