Coding & Development

ner-lstm is an open-source project that provides an implementation of Named Entity Recognition (NER) using multilayered bidirectional Long Short-Term Memory (LSTM) networks. This tool is based on the approach described in a research paper published at the ICON-16 conference. It leverages TensorFlow for its deep learning architecture and supports classification tasks for named entities in text corpora. The project includes functionalities for generating embedding models (Word2Vec, GloVe, RnnVec), preparing input data by resizing datasets and converting sentences to embeddings, and running the deep neural network. It has been tested on CoNNL 2003 NER Shared Task and the ICON-2013 Hindi NER dataset, demonstrating its applicability to both English and Hindi languages. The code is available on GitHub, making it accessible for developers and researchers interested in natural language processing.

LLM-Jailbreaks is a comprehensive GitHub repository dedicated to compiling and sharing various jailbreak prompts and techniques for large language models such as ChatGPT, Claude, and Llama. This resource is invaluable for developers, researchers, and security professionals looking to test the robustness and security of LLMs against adversarial inputs and prompt leaking. The repository includes detailed examples and methods like DeepSeek R1, Grok3, Gemini2.0, and different versions of DAN prompts, offering practical insights into bypassing typical AI safety measures. It serves as a community-driven platform for understanding and mitigating potential vulnerabilities in AI systems.

Pai-Megatron-Patch is an official deep learning training toolkit developed by Alibaba Cloud, designed for the large-scale training and prediction of Large Language Models (LLMs) and Vision-Language Models (VLMs) using the Megatron framework. It addresses the challenge of low training efficiency for models exceeding 10 billion parameters, effectively utilizing GPU computational power. The toolkit supports a wide range of commonly used LLMs, including Llama, Qwen, and DeepSeek, and incorporates accelerating techniques from Megatron-LM. Its design philosophy avoids invasive modifications to Megatron-LM's source code, instead providing expansions and improvements as patches. Key features include support for multiple LLMs, model weight conversion between Huggingface, Megatron, and Transformer Engine, FP8 training acceleration, and rich usage examples for pre-training, fine-tuning, evaluation, inference, and reinforcement learning.

PatrickStar is an open-source framework developed by Tencent that facilitates the parallel training of large language models (LLMs), particularly for Natural Language Processing (NLP) applications. It addresses the challenge of high hardware resource requirements for training PTMs by optimizing memory usage. Utilizing a chunk-based memory management system and heterogeneous training, PatrickStar efficiently leverages both CPU and GPU memory, allowing users to train significantly larger models with fewer GPUs. It has demonstrated the ability to train models like GPT3-175B on a relatively small GPU cluster, making advanced AI model training more accessible and cost-effective for a broader community.

The python-machine-learning-book-3rd-edition repository hosts the code examples for the third edition of the "Python Machine Learning" book. This resource is designed to complement the book, offering practical implementations of machine learning concepts using Python. It includes code for various chapters covering topics such as training algorithms, Scikit-Learn classifiers, data pre-processing, dimensionality reduction, model evaluation, ensemble learning, sentiment analysis, web application embedding, regression analysis, clustering, neural networks (from scratch and with TensorFlow), deep convolutional neural networks, recurrent neural networks, generative adversarial networks, and reinforcement learning. Users can explore these examples to deepen their understanding of the theoretical concepts presented in the book.

pyannote-whisper is an open-source tool designed for automatic speech recognition (ASR) and speaker diarization, leveraging the capabilities of Whisper for transcription and pyannote.audio for identifying and separating speakers. This tool allows users to process audio files to generate transcripts that include speaker labels and timestamps, making it ideal for analyzing multi-speaker conversations. It supports both command-line usage for quick processing and Python integration for more complex, programmatic workflows. The project provides clear examples for installation and usage, including how to integrate it into a Python script to diarize text and even generate meeting summaries using external LLMs like ChatGPT.

PyPOTS (pronounced "Pie Pots") is a comprehensive Python toolkit designed for machine and deep learning on partially-observed time series data. It addresses the common issue of missing values in real-world time series by offering a wide array of state-of-the-art neural network models for tasks such as imputation, classification, clustering, forecasting, and anomaly detection. The library is built to simplify complex data analysis, allowing engineers and researchers to focus on core problems rather than data preprocessing. PyPOTS integrates with an ecosystem of tools like TSDB for dataset loading, PyGrinder for simulating missing data patterns, and BenchPOTS for standardized performance evaluation. It also supports hyperparameter optimization for all neural network models, making it a robust solution for scientific analysis of incomplete industrial and irregularly-sampled multivariate time series.

Neurotime specializes in providing AI-powered solutions and marketing technologies designed to help businesses automate their processes, analyze complex data, and scale operations intelligently. The platform focuses on leveraging artificial intelligence to enhance efficiency and decision-making across various business functions. By offering advanced AI capabilities, Neurotime aims to empower companies to streamline their workflows, gain deeper insights from their data, and achieve sustainable growth. The solutions are tailored to support businesses in adapting to evolving market demands and optimizing their strategies through intelligent automation and data analysis.

stable-diffusion-webui-colab offers a web-based interface for Stable Diffusion, enabling users to generate images with a wide array of pre-trained models. The tool supports different versions of Stable Diffusion, including v1.4, v1.5, v2.0, and v2.1, along with specialized models like Waifu Diffusion, Midjourney v4, and various artistic styles. It provides options for training DreamBooth and LoRA models, as well as integrating ControlNet for enhanced image manipulation. The repository, while noted as outdated, serves as a comprehensive resource for exploring diverse diffusion models within a Google Colab environment.

Versatile Diffusion (VD) is presented as the first unified multi-flow multimodal diffusion framework, aiming for Universal Generative AI. It natively supports a range of generative tasks including image-to-text, image-variation, text-to-image, and text-variation. The framework is designed to be extensible, with future versions planned to support additional modalities such as speech, music, video, and 3D. VD's architecture comprises a VAE, a diffuser, and a context encoder within each flow, allowing it to handle specific tasks under various data and context types. It also facilitates advanced applications like semantic-style disentanglement and image-text dual-guided generation. The project provides a convenient WebUI for easy access to its features and is implemented in PyTorch, making it a valuable resource for researchers and developers in the field of diffusion models and multimodal generation.

The Udemy-prompt-engineering-course GitHub repository serves as a comprehensive companion for the Udemy Prompt Engineering Course. It is built around Jupyter notebooks, small example applications, prompt files, datasets, screenshots, and diagrams, offering a hands-on approach to learning prompt engineering. The repository covers a wide array of topics including OpenAI API workflows, advanced prompting techniques, retrieval, embeddings, RAG, agent design and orchestration, LangChain, LangGraph, evaluation, vision, and image generation. It is particularly strong in coding-heavy sections, providing practical examples for developers and teams looking for reference notebooks for LLM application development.

UniWorld is an open-source project from PKU-YuanGroup focused on advancing visual AI through high-resolution semantic encoders. It provides a unified framework for understanding, generation, and editing of visual content. The project includes UniWorld-OSP2.0 for VLM-enhanced image-to-video generation, UniWorld-V2 for image editing using diffusion models, and UniWorld-V1 for a broad range of visual tasks. All data, models, training code, and evaluation code are open-sourced, making it a valuable resource for AI researchers and computer vision engineers. The framework demonstrates excellent performance across various tasks, including subject consistency, background consistency, and aesthetic quality in video generation, and precise instruction execution in image editing.

The TensorFlow-Machine-Learning-Cookbook is a comprehensive code repository published by Packt, designed to accompany the TensorFlow Machine Learning Cookbook. It offers all the necessary project files to work through the book, enabling users to gain practical experience with TensorFlow. The resource covers fundamental concepts such as variables, matrices, and data sources, progressing to advanced topics like Linear Regression, neural networks, Convolutional Neural Networks (CNN), Recurrent Neural Networks (RNN), and Natural Language Processing (NLP). Each chapter's code is organized into folders, making it easy to follow along. It is compatible with Python 3 and requires libraries like TensorFlow, Numpy, Scikit-Learn, Requests, and Jupyter, running on Mac, Windows, and Linux without special hardware. This repository is ideal for those looking to deepen their understanding and application of Google's machine learning library.

Tensorflow-speech-recognition is an open-source project designed for speech recognition using Google's TensorFlow deep learning framework and sequence-to-sequence neural networks. It was developed as a replacement for caffe-speech-recognition. While the project is no longer actively maintained or up-to-date with the latest TensorFlow versions or state-of-the-art theory, it remains valuable for educational purposes. The repository provides various scripts for tasks like number classification, speaker classification, and speech-to-text, along with installation instructions for dependencies like pyaudio and portaudio. Users interested in modern speech recognition are advised to explore alternatives like Mozilla DeepSpeech or Whisper.

NLP-Projects is a comprehensive open-source repository dedicated to Natural Language Processing. It provides a wide array of concepts and practical scripts covering fundamental and advanced NLP topics. Users can explore implementations for word2vec, sentence2vec, machine reading comprehension, dialog systems, and text classification. The collection also delves into pretrained language models like XLNet, BERT, ELMo, and GPT, alongside sequence labeling, information retrieval, information extraction, knowledge graphs, text generation, and network embedding. It serves as a valuable resource for understanding and implementing various NLP techniques, with some sections offering Chinese notes for deeper insights.

ACG2vec (Anime Comics Games to vector) is an open-source AI tool dedicated to exploring deep learning applications within the ACG domain. It provides various functionalities including text semantic retrieval, image-to-image search, semantic image search, and image super-resolution. The project features several deep neural network models such as ACGVoc2vec for generating feature vectors from ACG-related text, DCLIP for fine-tuning CLIP models on ACG datasets, and Pix2Score for predicting image popularity and NSFW levels. It also includes Real-CUGAN's TensorFlow implementation for anime image super-resolution, making it a comprehensive playground for ACG and deep learning enthusiasts.

AI_Curriculum is an open-source repository offering a comprehensive collection of Deep Learning and Reinforcement Learning lectures from leading universities such as Stanford, MIT, and UC Berkeley. This resource is designed to support students and educators in the field of artificial intelligence by providing access to high-quality, structured learning materials. The curriculum covers various topics including Applied Machine Learning, Introduction to Deep Learning, CNNs for Visual Recognition, NLP with Deep Learning, Unsupervised Learning, Multi-Task and Meta Learning, and Deep Reinforcement Learning. Each section typically includes links to lecture videos, course websites, and sometimes GitHub notebooks, making it a valuable hub for self-paced learning and academic reference.

Algorithm_Interview_Notes-Chinese is an open-source GitHub repository offering extensive interview notes for various technical roles, including algorithm, deep learning, and natural language processing (NLP). The resource is designed to assist candidates preparing for job interviews in 2018, 2019, and during spring/autumn recruitment seasons. It covers a wide array of topics such as machine learning, deep learning, C, C++, and Python, alongside general computer science knowledge relevant to algorithm positions. The repository also compiles questions from numerous machine learning and deep learning interview experiences, providing a practical study guide. It explicitly excludes topics related to frontend, testing, Java, or Android development.

Stable Diffusion Arena is a Hugging Face Space designed for users to experiment with and compare various Stable Diffusion models. This tool facilitates the generation of high-quality images directly from text prompts. Users can select from different available models and fine-tune settings such as resolution and quality to achieve their desired visual output. It serves as an excellent platform for individuals interested in exploring the capabilities and performance differences between various AI image generation models within the Stable Diffusion ecosystem.

Bilateral AI is an Austrian Cluster of Excellence dedicated to advancing artificial intelligence by integrating symbolic and sub-symbolic AI. This project aims to overcome the limitations of current narrow AI systems, which are typically focused on specific tasks like object or speech recognition. By combining symbolic AI's logical rules with sub-symbolic AI's (like ChatGPT) data-driven learning, Bilateral AI seeks to develop 'Broad AI' capable of diverse applications and human-like reasoning. The initiative emphasizes creating AI that is not only fast and expandable but also safe, trustworthy, and understandable for everyday use. It involves cutting-edge research modules focusing on reasoning, learning, adaptability, and efficiency, and actively seeks to foster the next generation of AI researchers.

Chatbase-Alternative is an open-source project designed to help users create personalized chatbots for their websites. This tool allows you to train a chatbot on your specific website content, enabling it to instantly answer visitor questions. It positions itself as an alternative to commercial solutions like Chatbase, SiteGPT, and Dante AI, offering a cost-effective and customizable option for businesses and individuals. The project provides a tutorial and instructions for setting up the chatbot using Python, including requirements for OpenAI API key integration. It also mentions upcoming Replit and Streamlit versions for easier deployment, making it accessible for those looking to implement AI-powered customer support or information retrieval on their sites.

chatgpt-shell provides a multi-LLM Emacs comint shell, allowing developers to interact with a wide range of AI models including ChatGPT, Claude, DeepSeek, Gemini, Kagi, Ollama, Perplexity, and OpenRouter. It integrates seamlessly into the Emacs environment, offering a familiar shell experience with advanced features like a compose buffer for crafting detailed queries and a transient menu for quick access to common actions. Users can swap between different LLM providers, execute code snippets, and confirm inline modifications via diffs. The tool also supports vision experiments for image queries and offers integrations with Emacs org babel for evaluating code blocks. It is designed for developers who want to leverage AI within their Emacs workflow, providing flexibility and efficiency for various coding and text generation tasks.

ChatYuan-large-v2 is an open-source large language model developed by ClueAI, designed for dialogue in both Chinese and English. It builds upon its predecessor, ChatYuan-large-v1, by optimizing fine-tuning data, human feedback reinforcement learning, and chain-of-thought capabilities. This model is notable for its lightweight design, allowing inference on consumer-grade graphics cards, PCs, and even mobile phones (requiring as little as 400MB for INT4). Key enhancements include improved basic conversational and creative writing abilities, a new refusal-to-answer capability for dangerous questions, and added functionalities for code and table generation. It also boasts enhanced basic mathematical operations and an expanded maximum length from 1024 to 4096 tokens, alongside improved scenario simulation.

Awesome-LLM-Learning is a comprehensive open-source repository designed to guide individuals through the intricacies of Large Language Models (LLMs). It offers foundational knowledge in deep learning and natural language processing, essential for understanding LLMs. The resource delves into core LLM concepts, including training frameworks like Megatron-lm and DeepSpeed, parameter-efficient fine-tuning (PEFT), classic open-source LLMs, RLHF, CoT/ToT, and SFT training. Additionally, it covers LLM inference techniques such as Huggingface parameters and KVCache, and explores applications like LangChain. The repository also features a section dedicated to cutting-edge research, recommending relevant papers and blogs to keep learners updated with the latest advancements in the field. It's an invaluable resource for both newcomers and experienced professionals looking to deepen their understanding and practical skills in LLM development.