AI Trending

Project directory

Projects

Open-source AI projects tracked by the leaderboard, ordered by GitHub stars.

Vue 14 stars

Jannchie/chat-ui

Jannchie‘s Chat UI, for LLM Services

Overview: A generic frontend-only page for accessing various LLM APIs, designed by Jannchie. Key features: - Support for ChatGPT API. - Support for Open Router API. - Compatibility with all OpenAI-compatible platform APIs (Custom mode). - Support for chat history. - Displays reasoning process (Thinking). - Implemented text fade-in effect similar to Gemini. - Replica of Gemini-style chat UI. - Renders mathematical formulas, optimized for outputs from models like ChatGPT. - Supports code block copying and Markdown-formatted text copying. - Image Upload. Tech stack: Vue. Why it matters: This chat UI is designed to be a versatile frontend component for interacting with various language models, making it useful for developers and users alike. Repository: Jannchie/chat-ui Primary language: Vue GitHub stars: 14
Rust 14 stars

KMJ-007/VecPuff

vector database built on top of S3, inspired by turbopuffer

### Overview: VecPuff is a vector database built on top of Amazon S3, inspired by the turbopuffer architecture. It is designed for learning and experimentation with scalable vector search architectures. ### Key features: - **Scalable Vector Search**: Built on S3 for efficient storage and retrieval. - **Log-Structured Architecture**: Uses WAL, compaction, and ANN index for fast search. - **Local Cache**: Caches frequently accessed files for improved performance. - **HTTP Server**: Runs on `http://localhost:3000` by default. ### Tech stack: Rust ### Why it matters: VecPuff simplifies the process of experimenting with scalable vector search architectures, making it accessible for learning and development.
Jupyter Notebook 14 stars

LimDoHyeon/EEG-LLM

Fine-tuned LLM for electroencephalography(EEG) data classification

Overview: This project fine-tunes a large language model (LLM) for EEG data classification, utilizing four movements (left hand, right hand, tongue, foot) to predict them from EEG signals. The model achieves better performance through fine-tuning and demonstrates the potential of LLMs in supervised learning. Key features: - Fine-tuned LLM for EEG data classification - Utilizes LLM for classification operations - Trained on gpt-4o model for improved performance - Performance metrics: Accuracy, F1 score, ROC-AUC Tech stack: Python, openai, mne Why it matters: This project showcases the potential of LLMs as supervised learning models, highlighting their capabilities in EEG data classification.
C++ 14 stars

MarsPain/easy_llm.cpp

A minimal C++ framework for learning and understanding the LLM inference pipeline. The project aims to be readable, easy to learn, and easy to modify, while preserving the key steps of a full inference flow (config loading, weight loading, tokenizer, prefill/decode, sampling and decoding)

Overview: This project provides a minimal C++ framework for understanding and learning about the LLM inference pipeline. It includes key steps like config loading, weight loading, tokenizer, prefill/decode, sampling, and decoding, aiming for readability and modifiability. Key features: - Continuous batching service mode - Regression/invariant tests - Strong architecture/key/shape validation - CPU-first by default Tech stack: C++ Why it matters: It simplifies the learning process by providing a clear and editable pipeline, making it accessible for coursework, research, and self-study without focusing on peak performance. Repository: MarsPain/easy_llm.cpp GitHub stars: 14 GitHub description: A minimal C++ framework for learning and understanding the LLM inference pipeline.
HTML 14 stars

Mattbusel/llm-cpp

Add ChatGPT or Claude to C++ with one file.

Overview: This single-header C++17 library provides a set of 26 self-contained headers for integrating LLM (Large Language Model) features into C++ applications. These headers include streaming, retries, caching, cost estimates, RAG (Read-After-Write), structured JSON output, tool-calling agents, and more. The library is designed to be drop-in compatible with OpenAI and Anthropic APIs, making it suitable for developers shipping C++ applications without the need for a Python SDK or package manager. Key features: - **Streaming**: Support for real-time responses. - **Retries**: Automatic retry mechanisms. - **Caching**: Efficient caching of responses. - **Cost Estimation**: Real-time cost estimates. - **RAG**: Retrieval-Augmented Generation. - **Structured JSON Output**: Output in a structured format. - **Tool-Calling Agents**: Agents that can call external tools. - **Scrubbing PII and Keys**: Protection against exposing personal information. - **Cheaper Model Routing**: Optimized model routing for lower cost. - **Logging and Trace**: Detailed logging and tracing capabilities. - **Batch Prompts**: Support for batch processing. - **Mocking**: Mocking APIs for testing purposes. - **Images and Audio**: Support for image and audio processing. Tech stack: The library is built using C++17 and relies on the C++ standard library for most functionality. It also requires libcurl for API calls to OpenAI and Anthropic. Why it matters: This library simplifies integrating LLM features into C++ applications,
Python 14 stars

MohamedMostafa259/pif-multimodal-rag

A modular, multilingual, and multimodal Retrieval-Augmented Generation (RAG) system tailored for the financial analysis of Public Investment Fund (PIF) annual reports.

Overview: A modular, multilingual, and multimodal Retrieval-Augmented Generation (RAG) system tailored for the financial analysis of Public Investment Fund (PIF) annual reports. Key features: - Modular FastAPI backend - Vector database integration (Qdrant) - PDF ingestion and caching - Web frontend for interactive analysis - Prometheus metrics - Dockerized deployment Tech stack: Python Why it matters: This framework extends the design principles of M3DocRAG, specifically adapted for financial document understanding in both Arabic and English, making it a valuable tool for financial analysts and researchers. Repository: MohamedMostafa259/pif-multimodal-rag GitHub stars: 14
Python 14 stars

NFDI4BIOIMAGE/omero-vitessce

OMERO.web plugin for the Vitessce multimodal data viewer.

### Overview: OMERO.vitessce is an OMERO.web plugin designed for the Vitessce multimodal data viewer, providing enhanced functionality and user experience. ### Key features: - **Right-Hand Side Panel**: Allows users to select predefined configurations or generate custom ones for the Vitessce viewer. - **Open With**: Enables users to open the Vitessce viewer directly from the OMERO.web interface. - **Autogenerated Configurations**: Automatically generates and attaches configuration files for the viewer, simplifying setup. ### Tech stack: Python ### Why it matters: OMERO.vitessce offers a more intuitive and efficient way to interact with the Vitessce viewer, enhancing the overall user experience and making it easier to manage multimodal data.
Python 14 stars

NooBlock/astrbot-rag-kb

中文互联网梗 / 黑话 RAG 语料库：5 个分层知识库（4544 条 chunk）+ AstrBot 导入工具 + 使用 wiki。

Overview: 这个项目提供了一个中文互联网梗、黑话和社区语气的RAG语料库，包括5个层次的中文互联网语料库，用于增强中文聊天机器人的对话能力。 Key features: - 5个层次的中文互联网语料库 - 4544个chunk的中文互联网语料 - 用于AstrBot知识库的chunks.jsonl格式 - 验证环境：AstrBot v4.29.3 + Qwen/Qwen3-Embedding-8B - 语料来源：公开数据集/网页清洗，原始抓取文件不在本仓库发布 - 第三方语料的许可状态多为未声明/仅限自用，需自行核对上游许可 - 分层建库，避免语义混杂 - 语料校验仅在v4.29.3的接口验证过 Tech stack: Python Why it matters: 本项目通过提供中文互联网梗、黑话和社区语气的RAG语料库，增强了中文聊天机器人的对话能力，使其能够更好地理解和回应中文网络上的流行梗和黑话，同时保持社区风格的对话方式。
Python 14 stars

PRITHIVSAKTHIUR/Flux-LoRA-DLC

Flux-LoRA-DLC is an experimental, advanced image generation and image-to-image styling platform built on top of the state-of-the-art black-forest-labs/FLUX.1-dev text-to-image foundation model. The suite includes an integrated multi-LoRA routing manager packed with a gallery collection of over 250 highly distinct aesthetic DLC adapters.

Overview: Flux-LoRA-DLC is an experimental, advanced image generation and image-to-image styling platform built on top of the state-of-the-art `black-forest-labs/FLUX.1-dev` text-to-image foundation model. The suite includes an integrated multi-LoRA routing manager packed with a gallery collection of over 250 highly distinct aesthetic DLC adapters. Key features: - Massive 250+ LoRA Library: Instantly browse and hot-swap through an extensive pre-configured array of artistic, cinematic, cartoon, and highly realistic LoRA adapters via a single interactive grid. - On-the-Fly External Mounting: Dynamically download and load any public FLUX-compatible adapter weights directly during active sessions by pasting its Hugging Face repository slug. - Latent Denoising Previews: Overrides standard generation pipelines with a custom rendering loop that hooks directly into the discrete flow-matching Euler scheduler to stream step-by-step image passes. - Image-to-Image Conditioning: Supports structural image-to-image transformations accompanied by a granular denoise strength factor to precisely balance source image influence with prompt weight. - Sleek Steel Blue Theme: Wrapped in an intuitive, responsive dashboard styled with deep gray and steel blue colors, incorporating real-time mathematical progress containers. Tech stack: Python Why it matters: Flux-LoRA-DLC offers creators an ultra-flexible playground for mapping advanced latent transformations, providing an open-ended custom repository lookup field for live third-party Hugging Face LoRA mounting, and featuring an open-ended custom repository lookup field for live third-party Hugging
Python 14 stars

ReSerendipity/TTS-MultiModel

自托管多引擎语音合成平台：整合六种本地 TTS 引擎，另有 Edge-TTS 在线兜底，支持声音设计、克隆与多角色配音。 | A self-hosted multi-engine text-to-speech platform for voice design, voice cloning, and multi-character dubbing, with an online fallback.

Overview: TTS MultiModel 是一个开源的多引擎语音合成平台，支持六种本地引擎和一个在线引擎，提供声音克隆、声音设计、剧本配音和流式生成等功能，适用于多种语言和界面需求。 Key features: - 六引擎统一编排，支持灵活切换 - 声音克隆：零样本克隆，支持参考音频和转写文本 - 声音设计：直接用自然语言描述音色生成语音 - 剧本配音：多角色对话自动分配说话人，支持批量合成 - 流式与批量处理：长文本 SSE输出，断点续跑支持 - 多后端与多语言界面：支持 NVIDIA CUDA、Apple MPS 和 CPU Tech stack: Python Why it matters: 该平台简化了多引擎语音合成的复杂性，提供了一站式的解决方案，适用于多种语言和界面需求，特别适合开发和教育场景。
Python 14 stars

Siddhant-K-code/LLMTraceFX

Evidence-first tools to measure, verify, and optimize LLM inference across local GPUs and OpenAI-compatible APIs.

### Overview: LLMTraceFX is an evidence-first inference toolkit designed to measure, verify, and optimize Large Language Model (LLM) inference across local GPUs and OpenAI-compatible APIs. It ensures that the model's output is correct and that optimizations are only applied when supported by the evidence. ### Key features: - **Measurement and Verification**: Collects and verifies model output against deterministic workloads. - **Configuration Optimization**: Recommends configurations only when they are supported by the evidence. - **Evidence-Based Approach**: Uses a canonical schema to collect and verify measurements. ### Tech Stack: - Python ### Why it Matters: LLMTraceFX provides a robust and evidence-driven approach to measuring and optimizing LLM inference, ensuring that only verified and supported optimizations are applied. This methodology helps in identifying and mitigating bottlenecks in inference, leading to more efficient and accurate model performance. ### Repository: Siddhant-K-code/LLMTraceFX - Primary language: Python - GitHub stars: 14 - GitHub description: Evidence-first tools to measure, verify, and optimize LLM inference across local GPUs and OpenAI-compatible APIs. - README excerpt: Measures what happened, verifies it was correct, and optimizes only what the evidence supports. It includes a synthetic example report preview and a modal runbook. ### Research / Publication: - **Paper**: Understanding GPU-Level Bottlenecks in Large Language Model Inference - **Authors**: Shubhanshu Kushwaha, Mamata Samal, and Siddhant Khare - **Publication**: Proceedings
Python 14 stars

SnowNation101/Nyx

Code for the paper [SIGIR'26]“Towards Mixed-Modal Retrieval for Universal Retrieval-Augmented Generation”

Overview: This repository contains the official implementation of the paper "Towards Mixed-Modal Retrieval for Universal Retrieval-Augmented Generation" by Chenghao Zhang, Guanting Dong, Xinyu Yang, and Zhicheng Dou. The project focuses on developing a unified mixed-modal retriever called Nyx, which is designed for Universal Retrieval-Augmented Generation (URAG) scenarios. Key features: - Nyx: A unified mixed-modal retriever tailored for URAG scenarios. - NyxQA: A large-scale mixed-modal QA dataset constructed for the project. Tech stack: Python. Why it matters: Nyx and NyxQA provide a comprehensive solution for mixed-modal retrieval tasks, demonstrating strong performance across both text-only RAG benchmarks and vision-language URAG tasks. Repository: GitHub repository: https://github.com/SnowNation101/Nyx Primary language: Python. GitHub stars: 14.
Jupyter Notebook 14 stars

TheRealFREDP3D/Ollama-Colab

Run Ollama on Google Colab with a public HTTPS endpoint via Cloudflare Tunnel or ngrok.

Overview: This project provides a public HTTPS endpoint for the Ollama model, accessible via Google Colab, enabling users to run Ollama on a free GPU instance and access the model from anywhere. Key features: - Installs Ollama on a Colab GPU instance - Interactive model browser with live library scraping and static fallback - GPU-fit filtering (only shows models that fit the detected VRAM) - Exposes the Ollama API over a public HTTPS tunnel - Supports the standard Ollama API and the Anthropic-compatible `/v1/messages` endpoint - Tunnel readiness polling, process registry, and clean shutdown Tech stack: Jupyter Notebook Why it matters: This project simplifies the process of using Ollama on a free GPU instance, making it accessible to a wider audience, and provides a reliable and secure way to interact with the model from anywhere. Repository: TheRealFREDP3D/Ollama-Colab GitHub stars: 13 GitHub description: Run Ollama on Google Colab with a public HTTPS endpoint via Cloudflare Tunnel or ngrok.
Python 14 stars

Thisisus7/ING-VP

An Interactive Game-based Vision Planning benchmark

Overview: ING-VP is an interactive game-based vision planning benchmark designed to evaluate the multi-step reasoning and spatial imagination capabilities of Multimodal Large Language Models (MLLLMs). Key features: - 6 distinct games - 300 levels per game - 6 unique configurations per game - Multiple comparison settings (image-only vs. text-only, single-step vs. multi-step, with-history vs. without-history) Tech stack: Python Why it matters: This benchmark reveals that current MLLMs struggle with spatial imagination and multi-step planning, offering insights into the capability requirements for MLLMs.
Python 14 stars

VirtuosoResearch/Generalization-in-graph-neural-networks

Measuring generalization properties of graph neural networks

Overview: We provide a toolset for evaluating Hessian-based measurements of graph neural networks, focusing on their correlation with observed generalization gaps in fine-tuned models. This approach aims to better understand and measure the generalization properties of these neural network architectures. Key features: - `compute_hessian_spectra.py`: Computes the trace and eigenvalues of the loss's Hessian matrix for each layer. - `compute_hessian_norms.py`: Computes the Hessian-based vector product. Tech stack: Python Why it matters: This method offers a more direct and interpretable way to assess the generalization capabilities of graph neural networks, potentially leading to more accurate and meaningful insights into their performance. Repository: VirtuosoResearch/Generalization-in-graph-neural-networks GitHub stars: 14 GitHub description: Measuring generalization properties of graph neural networks
JavaScript 14 stars

WaitWut/krea2-structured-prompt

Prompt builder for Krea-2 in ComfyUI. Fill structured fields and get flowing prose in the format the model expects, not a keyword list.

### Overview: This project is a prompt builder for Krea-2 in ComfyUI, designed to generate structured prose that closely mimics the style of the Krea-2 model. It converts structured fields into a flowing two-paragraph prompt, ensuring the text reads naturally as intended by the model. ### Key Features: - **Structured Fields → Krea2-native Prose**: Fill in boxes to get a clean, two-paragraph prompt in Krea-2's expected format. - **Dynamic Character List**: Add up to 5 characters, each with appearance, pose/action, and clothing/props sub-fields. Drag to reorder, trash to remove. - **Soft Nudge, Not a Wall**: Past 5 characters, a subtle notice appears indicating Krea-2's tendency to drop fine-grained attributes. No blocking. - **Preset Pickers**: Each fixed field and character sub-field has a dropdown of 5 curated examples. Pick one to fill the box. - **Confirm Before Clobbering**: If you've typed your own text, applying a preset asks first; if the box is empty or still holds a known preset, it just fills silently. - **Prompt JSON Debug Output**: A second output dumps every raw field value as JSON, useful for debugging or downstream use. - **Zero Dependencies**: Pure standard-library string assembly, no extra Python dependencies. ### Tech Stack: JavaScript ### Why It Matters: This node allows developers to easily create structured prompts for Krea-2, ensuring the generated text reads naturally and closely follows the model's
Unknown 14 stars

Xinyi2016/FInstruct

Overview: FInstruct is a curated dataset of instruction-following data extracted from recent research articles by MSCI, S&P, and arXiv. It was generated using the HuggingChat model and is currently being actively curated by the creator, Xinyi2016. Key features: - Contains instruction-following data from MSCI, S&P, and arXiv research articles. - Generated using the HuggingChat model for financial instruction. - Quality labels available for 0 (Not curated), 1 (Should be excluded), and 2 (Pass basic cleansing). Tech stack: Python, HuggingChat, and data extraction tools. Why it matters: FInstruct provides a valuable resource for researchers, academics, and industry professionals interested in exploring the latest research in quantitative finance, and can be used as a training data set for fine-tuning language models for financial analysis and decision-making.
Python 14 stars

YuvrajSingh-mist/SmolHub

A collection of my from-scratch implementation of various models deployed on HF Spaces

Overview: This repository contains my original implementation of various models deployed on HF Spaces, merged into a new project called "SmolHub". Key features: - Original models and code - Updated documentation and features Tech stack: Python Why it matters: This repository is a collection of my original models and code, providing a valuable resource for developers and researchers interested in these models. Repository: YuvrajSingh-mist/SmolHub Primary language: Python GitHub stars: 14 GitHub description: A collection of my from-scratch implementation of various models deployed on HF Spaces README excerpt:
TypeScript 14 stars

ZeR020/opencode-mem0

OpenCode plugin that gives coding agents persistent memory using local vector database

Overview: OpenCode-Mem0 is an OpenCode plugin that provides persistent memory using a local vector database, eliminating the need for cloud services. This plugin stores data locally, ensuring that all data remains on your machine. Key features: - Auto-capture of conversations for persistent memory - Ability to add, search, list, forget, and profile memories - Supports multiple memory providers (e.g., OpenAI, Anthropic, Google Gemini) - No cloud services required, making it ideal for developers and small teams Tech stack: TypeScript Why it matters: This plugin simplifies memory management for developers by providing a reliable, local solution without the need for cloud storage, making it ideal for projects requiring high performance and security. Repository: ZeR020/opencode-mem0 Primary language: TypeScript GitHub stars: 14 GitHub description: OpenCode plugin that gives coding agents persistent memory using a local vector database (SQLite + usearch). No cloud services required — all data stays on your machine.
Python 14 stars

agneet42/revision

[ECCV 2024] "REVISION: Rendering Tools Enable Spatial Fidelity in Vision-Language Models"

### Overview: REVISION: Rendering Tools Enable Spatial Fidelity in Vision-Language Models ### Key features: - **Spatial Fidelity Enhancement**: REVISION improves spatial fidelity in vision-language models by generating spatially accurate synthetic images from textual prompts. - **Extensive Asset Support**: The framework supports 100+ 3D assets, 11 spatial relationships, diverse camera perspectives, and backgrounds. - **Training-Free Improvements**: Leveraging images from REVISION as additional guidance consistently improves the spatial consistency of T2I models across all spatial relationships. ### Tech stack: Python ### Why it matters: Utilizing rendering-based frameworks is an effective approach for developing spatially-aware generative models, addressing the limitations of current vision-language models in reasoning over spatial relationships.