Research & Education

DankGPT is an AI-powered research assistant designed for students and academics to efficiently analyze complex content and multiple documents. Leveraging GPT-4o-mini, it provides instant insights and supports comprehensive academic research across various topics. Users can upload documents and ask questions, receiving answers with citations to the source material. The tool streamlines the process of breaking down intricate information, making it an invaluable aid for scholarly work. It offers a free plan with limited features and a Pro plan for unlimited documents and custom prompts, catering to different research needs.

Awesome-LLM-for-RecSys is an open-source GitHub repository that serves as a comprehensive survey and collection of papers and resources focused on the application of large language models (LLMs) in recommender systems. It categorizes research based on how LLMs are adapted within the recommendation pipeline, including LLM for Feature Engineering (user/item-level augmentation, instance generation) and LLM as Feature Encoder (representation enhancement, unified cross-domain recommendation). The repository is regularly updated with the latest research, including a survey paper accepted by ACM Transactions on Information Systems (TOIS). It's an invaluable resource for academics and industry professionals looking to stay current with advancements in LLM-enhanced recommendation technologies.

ApplyAce is Pakistan's online admission platform, designed to connect students from all over the world with universities and colleges. Its mission is to make education accessible to everyone, everywhere. The platform provides expert guidance and AI-powered tools to help students find programs at top universities, both Pakistani and international. Key features include university and program search, eligibility checks, scholarship finders, and application assistance tools such as Statement of Purpose (SOP) generators, recommendation letter generators, CV generators, and personalized email generators. ApplyAce aims to streamline the admission process, making it easier for students to explore options, compare programs, and submit applications efficiently.

chatgpt_system_prompt is a comprehensive, open-source repository offering a diverse collection of GPT system prompts and insights into prompt injection and leaking knowledge. This resource is invaluable for anyone looking to deepen their understanding of how to craft effective system prompts for ChatGPT and other AI products, as well as for creating custom GPTs. It serves as an educational tool, highlighting various prompt writing techniques and demonstrating prompt injection security measures. The project emphasizes knowledge sharing to enhance prompt writing skills and raise awareness about security vulnerabilities, noting how many GPT authors have improved their defenses by learning from the breakdowns presented in this collection. It's a practical guide for both learning and contributing to the evolving field of prompt engineering.

genei is an AI-powered research and summarization tool designed to enhance productivity for academic and professional users. It leverages a custom AI algorithm to instantly summarize articles, analyze research, and extract key information from documents and webpages. Users can store and organize content in customizable projects and folders, with automatic referencing capabilities. The tool also features a Chrome extension for on-the-fly summarization and saving of webpages. genei offers AI-powered question answering, multi-document summarization, and rephrasing/paraphrasing functionalities, making it a comprehensive solution for managing and understanding research materials more efficiently.

協助專業人士和學習者快速處理海量資料與資訊並利用適當AI工具的小助手 is a small AI assistant designed to help professionals and learners efficiently manage large amounts of data and information by leveraging appropriate AI tools. This Chrome extension facilitates quick copy and paste functionality, automatically including the source URL when highlighting text with the mouse. It also supports standard keyboard shortcuts (Ctrl+C and Ctrl+V) for pasting both text and screenshots. A key feature is the ability to set special tags like "Keyword" and "To check" within drafts, allowing users to add comments for annotation. This streamlines subsequent organization and verification processes, making it an invaluable tool for research, data compilation, and document preparation.

Hulu-Med is a transparent, open-source generalist model designed for holistic medical vision-language understanding. It unifies understanding across diverse modalities including medical text, 2D/3D images, and surgical videos. Built with a focus on transparency and accessibility, Hulu-Med achieves state-of-the-art performance on 30 medical benchmarks, trained entirely on public data. Key features include holistic multimodal understanding, a fully open-source pipeline, and efficient training. It supports 12 major anatomical systems and 14 medical imaging modalities, covering diverse downstream tasks like medical report generation and anomaly detection. The model is available in various parameter scales (4B to 235B) and is compatible with HuggingFace Transformers and vLLM for easier integration and faster inference.

MiBA (Meaningful Insights Biotech Analytics) is a healthcare AI technology company specializing in oncology. It offers AI-powered data solutions to enhance oncology innovation and improve patient outcomes. The platform provides remarkable visibility into oncology care through a comprehensive, geographically diverse network of providers, consolidating all data assets in real-time with nightly updates. Key offerings include real-time intelligence for clinical trials, business intelligence with a proprietary Market Basket solution, targeted education for providers, and custom data solutions. MiBA's advanced AI technology synthesizes multimodal data to uncover meaningful patterns and insights, ensuring high precision and recall for reliable intelligence.

llama3.np offers a pure NumPy implementation of the Llama 3 model, making it an excellent resource for researchers and developers interested in understanding the underlying architecture of large language models. The project was validated using the stories15M model trained by Andrej Karpathy, ensuring an accurate and reliable implementation. It provides a straightforward way to run the Llama 3 model using Python and NumPy, demonstrating the core mechanics without complex dependencies. This tool is particularly valuable for academic research and educational contexts, allowing for detailed exploration and experimentation with the Llama 3 model's components.

LION (Latent Point Diffusion Models for 3D Shape Generation) is an open-source project presented at NeurIPS 2022, offering a robust framework for generating 3D shapes. This tool leverages advanced diffusion models to create 3D point clouds, enabling researchers and developers to explore and innovate in the field of 3D content creation. It includes functionalities for training VAE and diffusion prior models, with options for conditioning inputs like CLIP image embeddings for tasks such as single-view reconstruction or text-to-shape generation. The project provides detailed installation instructions, demo scripts, and evaluation tools, making it a valuable resource for those working with 3D shape synthesis and analysis.

LightReasoner is an innovative open-source research tool that redefines how large language models (LLMs) acquire reasoning capabilities. It leverages small language models (SLMs) to strategically identify critical reasoning moments, allowing LLMs to focus their learning more efficiently. This approach achieves superior performance with remarkable token efficiency, reducing total training time by 90%, sampled problems by 80%, and tuned tokens by 99% compared to traditional Supervised Fine-Tuning (SFT). The framework consists of a three-stage process: critical step selection via Expert-Amateur KLD detection, contrastive supervision, and self-distillation. LightReasoner demonstrates that strategic token selection, rather than exhaustive training, is key to unlocking latent LLM reasoning potential, proving that smarter, not blindly harder, is the path to scalable AI improvement.

Learn_Machine_Learning_in_3_Months is an open-source GitHub repository offering a structured curriculum for individuals aiming to learn machine learning in three months. Curated by Siraj Raval, this resource provides a week-by-week breakdown of topics, including foundational subjects like linear algebra, calculus, and probability, alongside practical skills such as Python for data science and an introduction to TensorFlow. The curriculum progresses to deep learning concepts, recommending resources like Fast.AI and suggesting project ideas. It serves as a comprehensive guide for self-study, linking to various YouTube playlists, online courses, and additional GitHub repositories.

long-context-attention, also known as Unified Sequence Parallelism (USP) or Hybrid Sequence Parallelism, offers a novel approach to training and inference for long context Large Language Models (LLMs). This open-source project synergizes the strengths of DeepSpeed-Ulysses-Attention and Ring-Attention, addressing their individual limitations. Ulysses-Attention is sensitive to the number of attention heads and less suitable for GQA/MQA scenarios, while Ring-Attention can be less efficient in computation and communication. LongContextAttention provides a more general, versatile, and performant solution. It supports various FlashAttention versions (v2, v3) and can even run without FlashAttention for NPUs. The tool includes functionalities for setting process groups, extracting local tensors, and offers different ring implementation types like 'zigzag' and 'basic'. It has been verified in Megatron-LM and applied in several other projects, providing a robust solution for researchers and developers working with long context generative AI.

Magentic-UI is a research prototype of a human-centered AI agent designed to automate complex web and coding tasks that may require monitoring. Unlike black-box agents, the system reveals its plan before executions, lets users guide its actions, and requests approval for sensitive operations while browsing websites, executing code, and analyzing files. Key features include co-planning for collaborative plan creation, co-tasking for guiding execution, action guards for sensitive operations, and plan learning/retrieval to improve future automation. It supports integration with Microsoft's Fara-7B model and offers flexible configuration for various LLM clients like Azure OpenAI and Ollama, making it a versatile platform for studying human-agent interaction.

LookaheadDecoding is an open-source project designed to significantly accelerate Large Language Model (LLM) inference by breaking the traditional sequential dependency of token generation. This innovative approach utilizes a parallel decoding algorithm, eliminating the need for a draft model or a separate data store. Motivated by Jacobi decoding, LookaheadDecoding collects and caches n-grams from Jacobi iteration trajectories, enabling simultaneous processing of future tokens. The process is divided into a lookahead branch, which generates new n-grams within a defined window, and a verification branch, which validates promising candidates. This method has demonstrated substantial latency reductions, achieving speedups ranging from 1.5x to 2.3x on various datasets and models. The tool supports sampling and FlashAttention, and is implemented with an attention mask to maximize GPU parallel computing power, making it a valuable resource for optimizing LLM performance.

machine-learning-surveys is a comprehensive GitHub repository offering a curated list of surveys, tutorials, and books related to machine learning. This resource is organized by topic, making it easy for users to find relevant literature on areas such as Active Learning, Bioinformatics, Classification, Clustering, Computer Vision, Deep Learning, Natural Language Processing, Reinforcement Learning, and more. Each entry typically includes the title, authors, and page count, with some entries highlighted for their significance. It serves as an excellent starting point for students, researchers, and professionals looking to deepen their understanding or explore specific subfields within machine learning.

MedMNIST is a comprehensive collection of 18 standardized biomedical image datasets, designed for 2D and 3D classification tasks. It includes 12 datasets for 2D images and 6 for 3D images, with various size options such as MNIST-like 28x28, and larger 64x64, 128x128, and 224x224 for 2D, plus 64x64x64 for 3D. These datasets cover diverse data modalities, scales (from 100 to 100,000 samples), and tasks (binary/multi-class, ordinal regression, multi-label). MedMNIST aims to simplify biomedical image analysis for researchers by providing pre-processed data and standardized train-validation-test splits, making it user-friendly for machine learning algorithm development and comparison. It is particularly useful for educational purposes due to its accessibility and lack of prerequisite background knowledge.

BIFOLD, the Berlin Institute for the Foundations of Learning and Data, conducts groundbreaking foundational research in Big Data Management (DM) and Machine Learning (ML), as well as their intersection. The institute is dedicated to educating future talents and generating high-impact knowledge in these critical fields. BIFOLD actively engages in various research projects, publishes scientific papers, and contributes to open-source systems, tools, and data. It fosters collaboration among researchers, policymakers, and industry representatives, as evidenced by events like BIFOLD Day. The institute also offers a Graduate School with innovative PhD programs for both bachelor's and master's degree holders, aiming to advance the next generation of AI and data science experts.

R Discovery is an AI-powered platform designed to streamline academic research for students and researchers. It offers access to over 300 million research papers, including 40 million open-access articles, across 32,000 journals. Users receive personalized reading feeds based on their interests, ensuring they stay updated with the latest and most relevant academic research. Key features include an AI assistant for summaries and paper references, a Chat PDF tool for interactive questioning, and a literature review generator. The platform also supports audio papers, paper translations into 30+ languages, and institutional access to paywalled articles, making academic research reading a more efficient and accessible experience.

TXYZ is an AI-powered platform designed to integrate all paths to knowledge, assisting users with research, learning, and problem-solving. It offers distinct products for searching, reading, and writing, each leveraging AI to redefine knowledge discovery, revolutionize knowledge work, and streamline workflows. The platform provides AI assistance for reading, problem-solving, and writing, with features like file upload and analysis, access to a comprehensive library, and agentic workflows. TXYZ is trusted by researchers and offers an API for businesses and institutions to empower their operations with AI. It aims to enhance accuracy and reduce 'hallucination' by refining its retrieval algorithm and providing document location references.

DeepSeek-V3 is a powerful Mixture-of-Experts (MoE) language model featuring 671B total parameters, with 37B activated for each token, ensuring efficient inference and cost-effective training. Building on the DeepSeek-V2 architecture, it introduces an innovative auxiliary-loss-free strategy for load balancing and a multi-token prediction training objective for enhanced performance. The model was pre-trained on 14.8 trillion diverse tokens and further refined through Supervised Fine-Tuning and Reinforcement Learning. DeepSeek-V3 demonstrates superior performance against other open-source models and rivals top closed-source alternatives, particularly excelling in math and code tasks. It supports local deployment on various hardware and open-source community software, including SGLang, LMDeploy, and TensorRT-LLM, with options for FP8 and BF16 inference.

micro_diffusion is an open-source repository from Sony Research that provides a minimalistic implementation for training large-scale diffusion models from scratch with an extremely low budget. Utilizing only 37 million publicly available real and synthetic images, it can train a 1.16 billion parameter sparse transformer for approximately $1,890, achieving a strong FID score on the COCO dataset. The repository includes training code, dataset code, and pre-trained model checkpoints for off-the-shelf generation. It supports progressive training from low to high resolution and incorporates patch masking for performance optimization and reduced training time.

ML2021-Spring is an official GitHub repository for the Machine Learning 2021 Spring course taught by Hung-Yi Lee at National Taiwan University. This resource offers comprehensive materials for students and self-learners, including code and slides for 15 distinct homework assignments. The assignments cover a wide range of machine learning topics, from fundamental concepts like Regression and Classification to advanced areas such as CNNs, Transformers, GANs, BERT, Autoencoders, Reinforcement Learning, and Meta Learning. The repository also provides links to the course website and lecture videos available on Hung-Yi Lee's YouTube channel, making it a valuable, self-contained learning package.

DeepExplain offers a comprehensive framework for understanding the behavior of deep neural networks through various attribution methods. It enables researchers and practitioners to interpret existing models and benchmark new attribution techniques. The tool supports both gradient-based methods like Saliency maps, Gradient * Input, Integrated Gradients, DeepLIFT, and epsilon-LRP, as well as perturbation-based methods such as Occlusion and Shapley Value sampling. DeepExplain is compatible with TensorFlow (V1) and Keras with a TensorFlow backend, providing flexibility for different development environments. Its capabilities help in identifying which input features contribute most to a network's output, aiding in debugging and model transparency.