Projects

Overview: This Flutter plugin allows developers to integrate speech recognition functionality on iOS and Android platforms, providing a seamless user experience. Key features: - Basic Example - Sytody, a speech to todo app Tech stack: Dart Why it matters: This plugin simplifies the integration of speech recognition technology into Flutter applications, making it accessible for developers without deep knowledge of iOS and Android APIs. Repository: rxlabz/speech_recognition GitHub stars: 334 GitHub description: A Flutter plugin to use speech recognition on iOS & Android (Swift/Java)

### Overview: LLMInternSkill is a comprehensive evidence-bound resume toolkit designed for internship resumes, job search optimization, and Codex Skill workflow. This project aims to polish resumes, tailor job descriptions, audit project materials, and prepare for interviews with tailored responses. It ensures that every element of the resume can withstand potential interview questions. ### Key Features: 1. **Resume Polishing**: Enhances the clarity, technicality, and alignment with the role of the resume. 2. **Job Tailoring**: Determines the fit of the job description and reorders experiences for better alignment. 3. **Project Packaging**: Converts project materials into authentic, evidence-rich stories. 4. **Interview Prep**: Generates interview questions, mock answers, and evidence upgrade plans. 5. **Project Scout**: Suggests ways to improve project materials and prepare for interviews. ### Tech Stack: - **Markdown** for documentation and user interface. - **GitHub** for version control and collaboration. - **LLM Internship Resume Polishing** for resume polish. - **JD Tailoring** for job description customization. - **RAG (Retrieval-Augmented Generation)** for project material analysis. - **Agent** for project material packaging. - **Post-Training Project Packaging** for project materials. - **Pretraining Project Packaging** for project materials. - **Search Ranking** for resume and project materials. - **AIGC (Artificial Intelligence Generated Content)** for evidence-rich materials. - **Multimodal** for integrating various types of evidence. ### Why It Matters: - **Enhanced Clarity**:

Overview: 本项目是一个使用 Flask 和 Pyecharts 构建的新冠肺炎疫情数据可视化交互分析平台，旨在提供疫情数据获取、每日疫情地图、曲线图展示、数据统计分析、态势感知、确诊人数预测分析算法设计、NLP 舆情监测等服务。 Key features: - 疫情数据获取 - 每日疫情地图展示 - 病毒曲线图展示 - 数据统计分析 - 态势感知 - 确诊人数预测分析 - NLP 舆情监测 Tech stack: Flask + Pyecharts Why it matters: 本项目通过结合空间数据分析和交互式可视化技术，为疫情防控中的数据感知、预测和决策提供有效依据和指南，助力全球抗击新冠肺炎疫情。

Overview: OrkasVideoStudio is a tool that allows developers to turn their coding agents into video studios, enabling them to describe videos in plain language and produce high-quality, editable, and re-renderable video files. Key features: - Compose: Script → HTML motion graphics → mp4 - Edit: Cut / Join / Trim-silence / De-filler / Mix / Burn-in subtitles / Localize - Generate: Your own provider keys (BYO) - Auto: Routes to a single cross-modal plan.json for complex deliverables Tech stack: TypeScript Why it matters: OrkasVideoStudio simplifies the process of creating engaging and shareable videos by providing a structured, editable plan that can be easily edited and re-rendered, making the video production workflow more efficient and accessible.

### Overview: AFT (Agent File Tools) is a Rust binary that integrates with an agent's environment, providing a sensorimotor cortex, motor cortex, and brainstem functionalities. It enhances productivity by offering a fast, efficient, and symbol-aware IDE and OS. ### Key Features: - **Sensory Cortex**: Perceives structure in code, enabling quick navigation and search. - **Motor Cortex**: Facilitates editing and refactoring, backed by binary validation and formatting. - **Brainstem**: Keeps the agent's environment running smoothly with background tasks and health checks. ### Tech Stack: Rust ### Why It Matters: AFT increases productivity by reducing token usage and improving the agent's ability to perceive and act on code effectively. It integrates seamlessly with existing tools like OpenCode, Pi, and OMP, providing a comprehensive solution for coding agents.

### Overview: EzAudio is a high-quality text-to-audio generation model designed for real-world audio applications, offering efficient diffusion transformer capabilities. It combines high-quality audio synthesis with reduced computational demands, making it suitable for various audio processing tasks. ### Key Features: - Efficient Diffusion Transformer for Text-to-Audio Generation - Supports Real-Time Text-to-Audio Conversion - In-Depth Audio Editing and Inpainting Capabilities - Integrates with Gradio for Interactive Audio Editing - Compatible with Hugging Face Spaces for Model Management ### Tech Stack: Python ### Why It Matters: EzAudio simplifies complex audio generation tasks by leveraging efficient diffusion techniques, making it a valuable tool for developers and researchers in the field of audio synthesis and processing. Its real-time capabilities and ease of integration make it an attractive choice for various applications, including but not limited to, music production, sound design, and audio analysis. ### Repository: haidog-yaqub/EzAudio Stars: 333 Description: High-quality Text-to-Audio Generation with Efficient Diffusion Transformer ### Installation: ```bash git clone git@github.com:haidog-yaqub/EzAudio.git cd EzAudio pip install -r requirements.txt ``` ### Usage: ```python from api.ezaudio import EzAudio import torch import soundfile as sf # Load model device = 'cuda' if torch.cuda.is_available() else 'cpu' ezaudio = EzAudio(model_name='s3_xl', device=device)

Overview: The Issue Label Bot is a GitHub App powered by machine learning that automatically labels issues based on their content. It predicts whether an issue should be labeled as a feature request, bug, or question, and applies the appropriate label if the prediction is accurate. Key features: - Automatically labels issues based on their content - Predicts labels using machine learning - Available on the GitHub Marketplace Tech stack: Python Why it matters: The Issue Label Bot automates the labeling process on GitHub, making it easier for developers to manage and track issues effectively. Repository: machine-learning-apps/Issue-Label-Bot Primary language: SCSS GitHub stars: 333 GitHub description: Code For The Issue Label Bot, an App that automatically labels issues using machine learning, available on the GitHub Marketplace. This is also code for the blog article: "How to automate tasks on GitHub with machine learning for fun and profit" README excerpt: [](https://www.python.org/downloads/release/python-360/) [](https://opensource.org/licenses/MIT) [](https://github.com/marketplace/issue-label-bot) <!-- TOC depthFrom:1 depthTo:6

Overview: Humla is an open-source meeting notes app for macOS, designed to capture and transcribe audio from meetings, including both microphone and computer audio streams. It records audio, transcribes it accurately, separates speakers, and generates summaries that fuse your notes with the transcript, all while keeping your data and audio on your machine. Key features: - Records two audio streams at once - Transcribes accurately in multiple languages - Identifies speakers offline Tech stack: Rust, Tauri 2, Swift Why it matters: Humla offers a local-first, privacy-focused solution for capturing and transcribing meeting notes, ensuring your audio and data stay on your machine. It's designed to be used in both remote and in-person meetings, making it a valuable tool for anyone looking to improve their meeting productivity and ensure their notes are accurate and useful. Repository: michaelwilhelmsen/humla GitHub stars: 329 GitHub description: Open-source AI meeting notes for Mac. Records mic + system audio with no bot, transcribes on-device or via OpenAI / Deepgram / Groq, identifies speakers offline, and writes summaries that fuse your notes with the transcript. Ask your notes and get cited answers. Tauri 2 + Rust + Swift.

### Overview: This repository is designed to provide a simple and intuitive explanation of reinforcement learning and decision-making concepts, focusing on the practical application using Jupyter Notebooks. The goal is to make these complex topics accessible to beginners, fostering a deeper understanding through hands-on learning. ### Key Features: - Reinforcement Learning and Decision-Making tutorials explained in an intuitive manner. - Practical demonstrations using Jupyter Notebooks. - Installation instructions for both Git and Docker. - Detailed explanations of decision-making problems, solutions, and evaluation methods. - Exercises to improve understanding and application of concepts. ### Tech Stack: Jupyter Notebook ### Why It Matters: By providing a clear and intuitive guide to reinforcement learning and decision-making, this repository aims to demystify these complex concepts, making them more accessible to learners. The use of Jupyter Notebooks allows for interactive learning and hands-on experience, which is crucial for building a deeper understanding of the material. ### Repository: mimoralea/applied-reinforcement-learning ### GitHub Stars: 333 ### GitHub Description: Reinforcement Learning and Decision Making tutorials explained at an intuitive level and with Jupyter Notebooks.

Overview: Cherche is a neural search framework designed to streamline the development of end-to-end pipelines for efficient and effective semantic search. Key features: - Neural search pipeline combining retrievers and pre-trained language models for both retrievers and rankers. - Well-suited for offline semantic search with compatibility for batch computation. - Provides a comprehensive set of features including retrievers, rankers, pipelines, and examples. Tech stack: Python Why it matters: Cherche simplifies the development of neural search pipelines, making it easier to create efficient and effective semantic search solutions. Repository: raphaelsty/cherche GitHub stars: 333 GitHub description: Neural Search

### Overview: A collection of resources and papers on Vector Quantized Variational Autoencoder (VQ-VAE) and its applications. ### Key features: - **Image Generation**: Techniques for generating high-quality images using VQ-VAE. - **Speech & Audio**: Applications of VQ-VAE in speech and audio processing. - **Low-level Vision**: Methods for generating low-level visual features using VQ-VAE. - **Others**: Applications beyond image and speech processing, including applications in low-level vision and other domains. ### Tech stack: Unknown ### Why it matters: VQ-VAE is a powerful technique that can be applied across various domains, from generating high-quality images to improving speech and audio processing. Its ability to learn hierarchical representations makes it particularly useful in applications where context and structure are crucial.

Overview: This repository contains the official code for the paper "Improving 2D Feature Representations by 3D-Aware Fine-Tuning," presented at ECCV 2024. The project focuses on fine-tuning 2D feature representations using 3D-aware fine-tuning techniques. Key features: - Colab Notebook - Hugging Face demo - ScanNet++ preprocessing code - Feature Gaussian training code - Fine-tuning code - Evaluation code Tech stack: Python Why it matters: This project showcases advancements in 2D feature representation through innovative 3D-aware fine-tuning techniques, potentially improving the accuracy and effectiveness of 2D models in various applications. Repository: ywyue/FiT3D GitHub stars: 332 GitHub description: [ECCV 2024] Improving 2D Feature Representations by 3D-Aware Fine-Tuning

### Overview: AudioBench is a universal benchmark designed to evaluate Audio Large Language Models (AudioLLMs) across various tasks such as speech, audio-scene, and voice understanding on 50+ datasets. It serves as a valuable tool for researchers and developers to assess the performance of different models. ### Key Features: - **Universal Benchmark**: Evaluates AudioLLMs on 50+ datasets. - **50+ Datasets**: Supports a wide range of datasets including ASR, speech recognition, audio-scene understanding, and voice recognition. - **Model Compatibility**: Compatible with a variety of models, including Qwen2-Audio-7B-Instruct. - **Model Evaluation**: Supports model-as-judge evaluation using VLLM framework. - **Model Inference**: Runs inference on models like Whisper and Llama-3, providing evaluation metrics like WER (Word Error Rate). ### Tech Stack: - **Python**: The primary programming language used for development and evaluation. ### Why It Matters: AudioBench provides a standardized way to compare and evaluate the performance of different AudioLLMs across various tasks. This benchmark is crucial for researchers and developers to objectively assess model capabilities and advancements in the field of audio language models.

### Overview: Artificial General Intelligence (AGI) is the intelligence of a machine that could successfully perform any intellectual task that a human being can. It is a primary goal of some artificial intelligence research and a common topic in science fiction and future studies. AGI is also referred to as "strong AI", "full AI", or as the ability of a machine to perform "general intelligent action". ### Key features: - **Instant communication chat channels**: Channels for real-time interaction and discussion. - **Articles**: Articles on various aspects of AGI, including its concept, history, and future prospects. - **Books**: Books exploring the idea of AGI, including "Life 3.0" and "The Society of Mind". - **Scientific Papers**: Papers on the latest research in AGI. - **Podcasts**: Podcasts discussing the topic of AGI. - **Courses**: Courses on AGI offered by educational institutions. - **Videos**: Videos explaining AGI concepts. - **Organizations / Projects**: Organizations and projects related to AGI. - **Forums / News / Social**: Forums, news, and social media platforms discussing AGI. - **Experts on Twitter**: Experts in the field of AGI sharing their insights. - **Competitions**: Competitions related to AGI. - **Movies**: Movies featuring AGI themes. - **TV Shows**: TV shows related to AGI. - **Anime**: Anime featuring AGI concepts. - **Education Plans**: Education plans and resources for learning about AGI. - **More resources**: Additional resources

### Overview: This project is an unofficial PyTorch implementation of ELECTRA, a pre-trained text encoder model designed by Kevin Clark, Minh-Thang Luong, Quoc V. Le, and Christopher D. Manning. The implementation is based on the original paper "[ELECTRA: Pre-training Text Encoders as Discriminators Rather Than Generators](https://arxiv.org/abs/2003.10555)" and successfully replicates the results from the paper on the GLUE dataset. ### Key Features: - **Pretrained Model:** The model is pre-trained from scratch using the OpenWebText corpus. - **Replication of Results:** The implementation closely follows the original paper's results, including the GLUE dataset scores. - **Training Stability:** The model shows good training stability across different seeds and runs, with a mean standard deviation (SD) of 0.57 for the GLUE devset. - **Jupyter Notebooks:** Comes with Jupyter notebooks for exploring the code and inspecting the processed data. ### Tech Stack: - Python ### Why It Matters: This implementation is valuable for developers and researchers interested in understanding and replicating the performance of ELECTRA, a model known for its performance on text classification tasks. The close replication of the original results and the inclusion of Jupyter notebooks make it a useful resource for further analysis and experimentation. ### GitHub Stars: 332 ### Repository: - Repository: richarddwang/electra_pytorch - Primary Language: Python - GitHub Stars: 3

### Overview: Weaviate is a vector database designed for machine learning applications, providing efficient search capabilities for various datasets. ### Key Features: - Semantic search through a vectorized Wikipedia (SentenceBERT) with Weaviate's vector search engine. - Search through Facebook Research's PyTorch BigGraph Wikidata-dataset with Weaviate's vector search engine. - Demo dataset for News Publications with Contextionary. - Demo dataset for News Publications with Transformers, NER, Spellcheck and Q&A. - Easy example of a schema and how to upload it to Weaviate with the Python client. - Semantic search through a wine dataset with Weaviate's vector search engine. - Basic example to get started with Weaviate and semantic search with the Transformers module. - Super simple 5-step guide to get started with the Weaviate NLP modules. - Pure vector database example using Weaviate without vectorization module, using BERT transformer to vectorize text documents. - Text search with Weaviate using own vectors (obtained using SBERT, but any other model can also be used). - A basic and simple example using our own vectors (obtained using SBERT, but any other model can also be used) in Weaviate. - A demo notebook showing how to use Weaviate as DocumentStore in Haystack. - An image classification example made using image2vec-neural and flask to classify vegetable images. ### Tech Stack: Python, JavaScript, GraphQL, Bash, React, YAML, Docker, HTML, Markdown, Weaviate

Overview: xk-time 是一个时间转换、计算、格式化、解析、日历、cron表达式和时间自然语言处理等工具，使用 Java 8，线程安全，支持多种日期格式化模板，轻量级，无第三方依赖。 Key features: - 时间转换 - 时间计算 - 时间格式化 - 时间解析 - 日历 - 时间cron表达式 - 时间自然语言处理 Tech stack: Java 8 Why it matters: xk-time 提供了 Java 8 API 的时间类库，如 Instant、LocalDate、LocalDateTime、LocalTime、ZonedDateTime 等，这些类库是线程安全的，并且增加了更丰富的功能。此外，它将时间转换、计算、格式化、解析等功能按功能横向组织，方便使用。它还支持日历、节假日、农历、二十四节气、十二星座、十二生肖、十二时辰和日历等功能，方便使用。

### Overview This repository contains the code, data, and trained models for our paper "BRIO: Bringing Order to Abstractive Summarization" at ACL 2022. BRIO introduces a novel training paradigm for neural abstractive summarization by incorporating a contrastive learning component that helps abstractive models estimate the probability of their generated summaries more accurately. ### Key Features - **Contrastive Learning**: Encourages abstractive models to generate summaries that are more accurate and coherent. - **Efficient Training**: Utilizes the Huggingface Transformers library for faster and more efficient training. - **Data Preprocessing**: Provides preprocessed datasets for CNN/DailyMail, XSum, and NYT, including the necessary license and pre-processing instructions. - **Candidate Summaries Generation**: Automatically generates candidate summaries from pre-trained models for easy evaluation and comparison. ### Tech Stack - **Python**: The primary language used for development. - **Huggingface Transformers**: The library used for training and evaluation of abstractive models. ### Why It Matters BRIO introduces a significant improvement in abstractive summarization by addressing the limitations of current methods, particularly in generating coherent and accurate summaries. This approach can lead to better performance and more practical applications of neural summarization models. ### Repository Details - **Repository**: yixinL7/BRIO - **Stars**: 332 - **Description**: ACL 2022: BRIO: Bringing Order to Abstractive Summarization - **Language**: Python - **GitHub

Overview: PMetal is a high-performance framework for Apple Silicon that enables efficient training, fine-tuning, inference, and serving of large language models using Rust. It simplifies the process by managing command-line interfaces, terminal UIs, desktop applications, and more. Key features: - Fine-tuning with the full LoRA family - Training adapts the model the server runs - Packed weights stay packed - One job spec for every surface - Drafters trained on a Mac - Quantization you can budget Tech stack: Rust Why it matters: PMetal simplifies the process of training, fine-tuning, and serving large language models on Apple Silicon, making it easier and more efficient for developers to leverage the GPU and Neural Engine capabilities of the Mac.

### Overview: StarWhisper 是一个涵盖天文语言模型、光变曲线与光谱分析、望远镜自主观测等领域的综合性项目。它包括训练数据、权重文件、科研技能包和项目目录，旨在自动化天文观测流程并提供相关科研工具。 ### Key features: - **天文语言模型**: 收录天文语言模型代码、数据和模型。 - **光变曲线与光谱分析**: 包括 Kepler / K2 分类、稀疏光变、SN Clock 和恒星光谱分析。 - **望远镜自主观测**: 包含 NGSS、全天相机、Explore 和 SitianClaw 等项目。 ### Tech stack: Python。 ### Why it matters: StarWhisper 是一个集天文语言模型、光变曲线与光谱分析、望远镜自主观测等科研技能于一体的综合性项目，旨在自动化天文观测流程并提供相关科研工具，对天文研究具有重要意义。 Repository: Yu-Yang-Li/StarWhisper GitHub stars: 331 GitHub description: Astronomy LLMs, StarWhisper Telescope, Virtual Sitian, and astronomy research skills