AI Trending

Project directory

Projects

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

Unknown 30 stars

apimart-2dwad/flux-2-api-fr

API FLUX.2 : identifiant de modèle, tarif 0.024 $ par image (1MP), paramètres, exemples cURL et Python, endpoint compatible OpenAI.

Overview: This project is an API called "API FLUX.2" that provides access to various image models, including 1MP, 2MP, and 3MP resolutions, with a cost of $0.024 per image. The API is compatible with OpenAI and supports a single API key for multiple models. Key features: - Access to 300+ image models - 1MP, 2MP, and 3MP resolution options - Single API key for multiple models - Compatible with OpenAI - Supports asynchronous requests Tech stack: Node.js Why it matters: This API is useful for developers who need to quickly generate high-quality images for their projects without the need for multiple API keys or expensive subscriptions. The low cost per image makes it an attractive option for both small and large-scale applications. Repository: apimart-2dwad/flux-2-api-fr Primary language: Unknown GitHub stars: 30 GitHub description: API FLUX.2 : identifiant de modèle, tarif 0.024 $ par image (1MP), paramètres, exemples cURL et Python, endpoint compatible OpenAI.
Unknown 30 stars

apimart00-12/grok-imagine-2-api-cn

Grok Imagine 2.0 API 中文文档：模型 ID、每张 $0.015 计费、图像参数与调用示例，含 curl / Python 代码与 OpenAI 兼容接口说明。

Overview: This repository contains detailed documentation for the Grok Imagine 2.0 API, including model IDs, pricing, and example calls in both curl and Python. It also provides information about the API key and task ID system. Key features: - Model ID: grok-imagine-2.0-ext - Pricing: $0.015 per image, $0.0000 per upload, $0.015 per region-edit, $0.015 per layer - API endpoint: POST https://api.apimart.ai/v1/images/generations - Task ID system: GET /v1/tasks/{id} Tech stack: Python and OpenAI API Why it matters: This repository serves as a comprehensive guide for developers looking to integrate Grok Imagine 2.0 into their projects, providing accurate pricing information and practical examples for both curl and Python scripting.
Python 30 stars

b3b41020/free-multimodal-proxy

OpenAI-compatible reverse proxy for free multimodal AI APIs (chat / images / videos / audio / 3d)

### Overview: This project is an OpenAI-compatible reverse proxy designed specifically for free multimodal AI APIs. It allows developers to use InferencePort AI's free multimodal API without requiring registration or API keys, making it easy to integrate with existing tools. ### Key features: - **Text chat**: Supports OpenAI-compatible `chat/completions` with SSE streaming for reasoning chunks. - **Image generation**: 25+ models (Flux, GPT-Image, Seedream, Qwen-Image, Ideogram, Imagen, Wan, ...) returning `b64_json`. - **Video generation**: 19+ models (Wan, Kling, Vidu, Seedance, PixVerse, Flux, Hailuo, ...) with async job API for `POST /v1/videos` and sync API for `POST /v1/videos/generations`. - **Audio / 3D passthrough**: Upstream status dependent. - **106 models**: Served from a cached `/v1/models` listing with a TTL of 5 minutes. - **Optional `PROXY_TOKEN`**: Bearer auth for public deployments. - **Bounded retries**: Retries on transient upstream failures with a 429/5xx/network timeout. ### Tech stack: Python ### Why it matters: This project provides a straightforward surface for OpenAI-compatible APIs, making it easy to integrate with existing tools without requiring registration or API keys. The public API endpoint is designed to be equivalent to the app's "studio" mode, allowing developers to use the
C# 30 stars

buraksenyurt/ai-based-development-lecture

Yapay Zeka destekli yazılım geliştirme dersine ait notların ve örnek uygulamaların yer aldığı kod reposudur.

### Overview: Bu kod reposu, Yapay Zeka destekli yazılım geliştirme dersine ait notlar ve örnek uygulamaları içerir. Öğrenciler, yapay zeka araçları ve teknolojilerin yazılım geliştirme süreçlerine nasıl entegrasyon edilmesi gerektiğini öğrenir. ### Key features: - Önsöz - Yapay Zeka Alanındaki Anahtar Terimler - Tanışma ve `Hello World` Uygulamasının Geliştirilmesi - CV Bank Projesi için Prototip Geliştirme - Exception Handling, Debugging ve Docker Kullanımı - Bağımlılıkları Yönetmek ve Kod Kalitesini Ölçmek - Yazılım Çözümlerinde Testin Önemi - Yazılım Mimarileri ve Temel Seviyede Bir Örnek - Dağıtık Sistemler Hakkında Temel Bilgiler ve Basit Bir Senaryo Üzerinden İnceleme - RAG (Retrieval Augmented Generation) Yaklaşımı I - RAG (Retrieval Augmented Generation) Yaklaşımı II - MCP (Model Context Protocol) Kavramı ve MCP Server Yazılması - MCP Server'lar ile Çalışmak - Custom Agent ve Skill Yapıları ile Çalışmak - LoRA (Low-Rank Adaptation) ile Model Özelleştirme - Yapay Zeka Destekli Yaz
Python 30 stars

catalina17/XFlow

Generalized cross-modal NNs; new audiovisual benchmark (IEEE TNNLS 2019)

### Overview: XFlow is a novel cross-modal deep neural network architecture designed to enhance data flow between audio and visual modalities, thereby improving classification performance. This method is particularly effective in achieving state-of-the-art results on benchmark datasets, including AVletters, CUAVE, and Digits. ### Key Features: - **Cross-Modality Data Flow**: XFlow allows for the exchange of representations between audio and visual features, leveraging their distinct dimensions and non-trivial exchangeability. - **Improved Interpretability**: Derives more interpretable features compared to models that do not exchange representations. - **New Benchmark Dataset**: Digits, consisting of three data types extracted from videos of people saying the digits 0-9. ### Tech Stack: Python ### Why It Matters: XFlow represents a significant advancement in cross-modal learning, providing a novel method for performing cross-modality processing before features are learned from individual modalities. This approach effectively exploits correlations between audio and visual data, leading to superior performance on benchmark datasets.
Python 30 stars

chahuja/mix-stage

Official Repository for the paper Style Transfer for Co-Speech Gesture Animation: A Multi-Speaker Conditional-Mixture Approach published in ECCV 2020 (https://arxiv.org/abs/2007.12553)

### Overview This repository contains the official code and pre-trained models for the paper "Style Transfer for Co-Speech Gesture Animation: A Multi-Speaker Conditional-Mixture Approach" published at ECCV 2020. The dataset used in the paper is available at the provided dataset website. ### Key Features - **Training Code and Pre-trained Models**: Contains detailed information on how to train models from scratch and access pre-trained models. - **Dataset**: Links to the dataset repository for downloading the dataset files. - **Experiment Files**: Information on the experiment files used in the paper, including the dataset and model parameters. - **Inception Score for Pose Sequences**: Details on how to calculate the Inception Score for pose sequences. ### Tech Stack - Python ### Why It Matters This repository provides a comprehensive resource for researchers and developers interested in style transfer techniques for multi-speaker co-speech gesture animation. It serves as a valuable resource for understanding and implementing the proposed method in various applications. ### Repository Details - **Repository**: `chahuja/mix-stage` - **GitHub Stars**: 30 - **GitHub Description**: Official Repository for the paper "Style Transfer for Co-Speech Gesture Animation: A Multi-Speaker Conditional-Mixture Approach" published in ECCV 2020 - **Primary Language**: Python ### Additional Information - **Clone**: Clone the repository to the master branch using the provided command. - **Set up Environment**: Ensure you have Python installed and set up an environment with the necessary dependencies. - **Training**: To train
TypeScript 30 stars

codefromkarl/ContextAtlas

ContextAtlas — context infrastructure for AI coding agents: hybrid retrieval, project memory and retrieval observability via CLI, MCP server or embeddable library. Tree-sitter indexing, LanceDB vector search, FTS5 and token-aware context packing.

### Overview: ContextAtlas is an open-source context infrastructure designed for AI coding agents, providing hybrid code retrieval, project memory, and retrieval observability. It combines tree-sitter semantic chunking, LanceDB vector search, SQLite FTS5 full-text search, and token-aware context packing to deliver structured, high-quality code context to tools like Claude Code, Codex, and custom agent workflows. ### Key Features: - **Hybrid Retrieval**: Retrieves code snippets from various sources. - **Project Memory**: Stores project-specific data for quick access. - **MCP Server**: Provides a centralized memory management service. - **Retrieval Observability**: Tracks and improves retrieval performance over time. ### Tech Stack: - TypeScript - Node.js - MCP Server - Tree-sitter indexing - LanceDB vector search - FTS5 full-text search - Token-aware context packing ### Why It Matters: ContextAtlas addresses the practical engineering problem of finding the right code faster in large repositories, persisting project understanding, and improving retrieval and indexing quality over time. It is essential for developers building Claude Code workflows, MCP clients, or custom agent workflows. ### Repository: - **Repository**: codefromkarl/ContextAtlas - **GitHub Stars**: 30 - **GitHub Description**: ContextAtlas — context infrastructure for AI coding agents: hybrid retrieval, project memory and retrieval observability via CLI, MCP server or embeddable library. Tree-sitter indexing, LanceDB vector search, FTS5 and token-aware context packing. ### Installation: - **Installation**: Follow
Jupyter Notebook 30 stars

crux82/CLiC-it_2023_tutorial

This repository hosts materials from the CLiC-IT 2023 tutorial

Overview: This repository contains materials from the CLiC-IT 2023 tutorial, focusing on introducing Transformer-based architectures, fine-tuning Large Language Models (LLMs) on diverse datasets, and utilizing Low-Rank Adaptation (LoRA) for sustainable and efficient tuning on modest hardware. Key features: - Introduction to Transformer-based architectures - Fine-tuning of Large Language Models on diverse datasets - Utilization of Low-Rank Adaptation for sustainable and efficient tuning Tech stack: Jupyter Notebook Why it matters: This tutorial provides a comprehensive guide on how to effectively fine-tune Large Language Models using sustainable and efficient methods, making it valuable for developers and researchers interested in leveraging these models for various applications.
Python 30 stars

divelab/Sys2Bench

Sys2Bench is a benchmarking suite designed to evaluate reasoning and planning capabilities of large language models across algorithmic, logical, arithmetic, and common-sense reasoning tasks.

### Overview: Sys2Bench is a benchmark suite designed to evaluate the reasoning and planning capabilities of large language models across a variety of tasks, including algorithmic, logical, arithmetic, and common-sense reasoning. It serves as a comprehensive resource for researchers and developers to assess the performance of models in complex reasoning scenarios. ### Key Features: - **Arithmetic Reasoning**: Evaluates models' ability to solve mathematical problems. - **Logical Reasoning**: Tests models' capability to reason through logical statements and arguments. - **Common-Sense Reasoning**: Measures models' understanding of everyday situations and common sense knowledge. ### Tech Stack: - Python ### Why It Matters: Sys2Bench provides a standardized benchmark that helps researchers and developers evaluate the performance of large language models across different reasoning tasks. This benchmark is crucial for advancing the field of artificial intelligence and ensuring that models can effectively handle complex reasoning scenarios in various domains. ### Repository: <div align="center"> <a href="https://github.com/divelab/Sys2Bench"> <img src="assets/sys2bench.png" alt="Logo" width="320" height="320"> </a> </div> ### GitHub Stars: 30 ### GitHub Description: Sys2Bench is a curated benchmark designed to evaluate the reasoning and planning abilities of large language models across algorithmic, logical, arithmetic, and common-sense reasoning tasks. It aims to provide a standardized and comprehensive resource for researchers and developers to assess the performance of models in complex reasoning scenarios. ### Announcements: -
Python 30 stars

finos/greenkey-asrtoolkit

A collection of useful tools for handling speech recognition data

Overview: GreenKey ASRToolkit is a collection of tools designed to simplify the workflow for building, customizing, and analyzing speech recognition models. Key features: - File formats support: STM, SRT, VTT, txt, and GreenKey JSON formatted transcripts. - `convert_transcript` tool for easy conversion among supported file formats. - `wer` tool for comparing reference and hypothesis transcripts. - `clean_formatting` script to standardize formatted text for ASR engine compatibility. - `split_audio_file` utility for splitting audio files based on valid segments from a transcript. - `prepare_audio_corpora` function to organize corpora into a target directory for training, testing, and development sets. Tech stack: Python Why it matters: This toolkit is essential for scientists, engineers, and other technologists involved in speech recognition, providing a streamlined approach to handling and analyzing speech data. Repository: finos/greenkey-asrtoolkit GitHub stars: 30 GitHub description: A collection of useful tools for handling speech recognition data NOTE! This project is archived due to lack of activity; you can still consume this software, although not advised, as it is not actively maintained. If you're interested to restore activity on this repository, please email help@finos.org
Python 30 stars

genia-dev/vibraniumdome

LLM Security Platform.

Overview: VibraniumDome is a cutting-edge open-source platform designed to secure LLM agents, providing comprehensive security measures against vulnerabilities such as prompt injections, model denial of service, and sensitive information disclosure. It includes a full-stack web application, a shields server, and an open-source, fine-grained policy management system. Key features: - Full-Stack Web-Application - Shields Server - OpenSearch for storage and analytics Tech stack: Python Why it matters: VibraniumDome empowers security teams to govern, audit, and control LLM usage effectively, ensuring compliance with enterprise-grade security practices. Repository: genia-dev/vibraniumdome GitHub stars: 30 GitHub description: LLM Security Platform.
Python 30 stars

git-disl/Lisa

This is the official code for the paper "Lazy Safety Alignment for Large Language Models against Harmful Fine-tuning" (NeurIPS2024)

### Overview: Lisa is a fine-tuning stage defense against the threat of harmful fine-tuning, designed to protect large language models from the risk of fine-tuning with harmful data. ### Key features: - Customized trainer implementation on top of HuggingFace Trainer. - Adds a line of code to switch dataset/model based on steps number for Bi-state optimization. - Adds regularization to loss function to introduce a proximal term towards consensus. - Provides package requirements for installation via `conda` and `pip`. - Includes scripts for preparing fine-tuning data, running experiments, and reproducibility with Slurm. - Supports fine-tuning with 10% of harmful data from SST2 dataset, and comparison with SFT. ### Tech stack: Python ### Why it matters: Lisa provides a robust defense mechanism against harmful fine-tuning, ensuring the safety and integrity of large language models.
Go 30 stars

gptscript-ai/knowledge

Knowledge for GPTScript

### Overview: A Knowledge Tool for GPTScript and GPTStudio, designed to enhance the development process by providing comprehensive documentation and insights. ### Key Features: - **Ingestion**: Easily import and manage documentation files like `.md`, `.txt`, `.docx`, `.pdf`, `.html`, `.rtf`, `.csv`, and `.ipynb`. - **Retrieval**: Retrieve information from the imported documents for quick access. - **Dataset Management**: Create, delete, and manage datasets for organization. - **Server Mode**: A server mode for enhanced functionality and remote access. ### Tech Stack: Go ### Why It Matters: This tool streamlines the documentation process, making it easier for developers to find and reference information quickly, improving productivity and efficiency in GPTScript projects.
HTML 30 stars

habitual69/speakify

Speakify is a web application that uses Edge TTS to convert text to speech using a variety of voices.

Overview: Speakify is a web application that converts text to speech using Edge TTS, featuring a variety of voices. Key features: - Text-to-speech conversion - Multiple voice options Tech stack: Python, HTML, JavaScript Why it matters: Speakify simplifies text-to-speech conversion, making it accessible and user-friendly for developers and content creators. Repository: habitual69/speakify GitHub stars: 30
Go 30 stars

hexxla/hexxladb

A hex-native embedded database for Go. Morton keys, native ring scans, seams (contradictions), MVCC, provenance, and optional HNSW embeddings. Built from scratch for lattice-addressed memory.

Overview: HexxlaDB is an embedded Go library designed for structured, spatially-organized records, providing features like Morton keys, native ring scans, seams (contradictions), MVCC, provenance, and optional HNSW embeddings. It is built from scratch for lattice-addressed memory. Key features: - Morton keys - Native ring scans - Seams (contradictions) - MVCC (Multi-Version Concurrency Control) - Provenance tracking - Optional HNSW embeddings Tech stack: Go Why it matters: HexxlaDB simplifies the development of spatially-organized applications by providing efficient and reliable database capabilities, including provenance tracking, conflict resolution, and vector search, all within a single binary package with no network dependencies or daemons.
Python 30 stars

hrishi-008/SummarAI

A tool for summarizing search results and website content using FAISS, LLMs, and the Retrieval-Augmented Generation (RAG) technique.

Overview: SummarAI is a tool designed to automate the process of summarizing search results and website content using FAISS, LLMs, and the Retrieval-Augmented Generation (RAG) technique. This application leverages the RAG technique to provide concise and context-aware summaries based on user queries. Key features: - Google Search Scraping - Website Content Extraction - FAISS Database Integration - Content Summarization with RAG - Streamlit Interface Tech stack: Python Why it matters: SummarAI streamlines the process of summarizing search results and website content, making it easier for users to quickly grasp the key points of their queries and quickly find relevant information.
Python 30 stars

huzongxiang/sowa

Overview: Soldier-Offier Window self-Attention (SOWA) is a novel method for zero/few-shot anomaly detection that leverages the CLIP model and learnable prompts to enhance hierarchical feature processing within a Soldier-Offier Window self-Attention framework. This approach demonstrates superior performance across various benchmark datasets compared to existing techniques. Key features: - Incorporates a Soldier-Offier Window self-Attention mechanism - Utilizes CLIP model for feature extraction - Implements learnable prompts for enhanced feature processing - Tested on five benchmark datasets Tech stack: Python Why it matters: SOWA offers a scalable and effective solution for anomaly detection in industrial manufacturing, particularly in scenarios where traditional methods rely on extensive normal datasets and custom models. By leveraging hierarchical features and zero/few-shot learning, SOWA significantly outperforms existing techniques, making it a valuable tool for improving the accuracy and efficiency of anomaly detection in various applications.
Python 30 stars

jiwidi/DeepSpeech-pytorch

Pytorch implementation for DeepSpeech 2.0

### Overview: DeepSpeech-pytorch is a PyTorch implementation of DeepSpeech 2.0, a state-of-the-art end-to-end speech recognition model. This project aims to provide a practical and efficient solution for speech recognition tasks using PyTorch. ### Key features: - PyTorch implementation of DeepSpeech 2.0 - End-to-end speech recognition model - Supports training and testing on various datasets - Tensorboard logs for model monitoring ### Tech stack: PyTorch ### Why it matters: This project provides a practical and efficient solution for speech recognition tasks using PyTorch, making it accessible and easy to integrate into existing projects.
Python 30 stars

kaushiknishchay/ComfyUI-Qwen3-ASR

ComfyUI nodes for Qwen3-ASR (0.6B/1.7B) and ForcedAligner. Supports high-accuracy ASR and language identification for 52 languages/dialects, including 22 Chinese dialects and various English accents. Features word-level timestamps, long audio transcription, and VRAM-optimized inference.

### Overview: ComfyUI Qwen3 ASR is a high-performance extension for the Qwen3-ASR model family, providing precise word-level timestamps, multilingual support, and automatic language detection. It is optimized for high-accuracy ASR and language identification across 52 languages and dialects, including 22 Chinese dialects and various English accents. ### Key Features: - **High Accuracy**: Supports Qwen3-ASR 0.6B and 1.7B models. - **Multilingual**: Supports 52 languages and dialects with automatic language detection. - **Word-Level Timestamps**: Optional integration with Qwen3-ForcedAligner-0.6B. - **Flexible Precision**: Support for `bf16`, `fp16`, and `fp32` to balance VRAM and speed. - **Automatic Resampling**: Internally handles audio resampling to 16kHz for optimal model performance. - **FlashAttention 2**: Integrated support for FlashAttention 2 to significantly reduce VRAM usage and accelerate inference. ### Tech Stack: Python ### Why it Matters: This extension leverages the latest Qwen3-ASR and Qwen3-ForcedAligner models, providing high-accuracy speech-to-text transcription, language identification, and precise word-level timestamps. The integration with FlashAttention 2 ensures faster inference and lower VRAM usage, making it ideal for developers and AI enthusiasts looking to enhance their applications with advanced speech processing capabilities.
Python 30 stars

kgnlp/allophant

A multilingual phoneme recognizer capable of generalizing zero-shot to unseen phoneme inventories.

Overview: Allophant is a multilingual phoneme recognizer designed for generalization to unseen phoneme inventories, showcasing zero-shot capabilities in INTERSPEECH 2023. Key features: - Multilingual support across 34 languages - Zero-shot generalization capability - Pre-trained checkpoints available on Hugging Face Tech stack: Python Why it matters: Allophant demonstrates advanced multilingual phoneme recognition and zero-shot generalization, making it a valuable tool for researchers and developers working on language modeling and cross-lingual applications. Repository: kgnlp/allophant Primary language: Python GitHub stars: 30 GitHub description: A multilingual phoneme recognizer capable of generalizing zero-shot to unseen phoneme inventories.