Coding & Development

ChatMLX is a modern, open-source, and high-performance chat application specifically designed for MacOS. It utilizes large language models (LLMs) and is built upon the powerful MLX framework, optimized for Apple silicon. The application supports multiple models, including Llama, OpenELM, Phi, Qwen, Starcoder, Cohere, and Gemma, offering users a diverse range of conversational options. A key differentiator is its ability to run LLMs locally, which significantly enhances user privacy and security. ChatMLX is multilingual, supporting 39 major App Store languages, making it accessible to a global audience. It's compatible with macOS 14.0 and later.

CodeGeeX4-ALL-9B is an open-source, multilingual code generation model designed for a wide range of AI software development tasks. Built upon the GLM-4-9B, it significantly enhances code generation capabilities, offering features like code completion, code interpretation, web search, and function calling. This versatile model also supports repository-level code Q&A, making it a comprehensive solution for developers. CodeGeeX4-ALL-9B has demonstrated highly competitive performance on public benchmarks such as BigCodeBench and NaturalCodeBench, often outperforming larger general-purpose models. It is recognized for achieving an optimal balance between inference speed and model performance, making it the most powerful code generation model with less than 10 billion parameters. The model is available on Ollama, Huggingface transformers, and vLLM, with extensions for VS Code and Jetbrains.

CodeT5 and CodeT5+ are open-source large language models developed by Salesforce Research, designed for advanced code understanding and generation tasks. These models function as an AI-powered coding assistant, significantly boosting developer productivity. Key capabilities include generating code from natural language descriptions, autocompleting entire functions based on a target function name, and summarizing code functions into natural language. The models are available on HuggingFace, with fine-tuned checkpoints for various downstream tasks, including multilingual code summarization. CodeT5 is released under the BSD-3 License, encouraging ethical use and user feedback on applications.

ChatALL is an open-source desktop application designed to streamline interactions with various large language models (LLMs). It enables users to send a single prompt to multiple AI bots simultaneously, including popular ones like ChatGPT, Bing Chat, Bard, Claude, and many others. This concurrent querying helps users compare responses and identify the best answers or creations from different LLMs. The tool is particularly useful for LLM gurus seeking optimal results, researchers comparing model strengths, and developers debugging prompts. It supports a wide range of AI bots, offering features like quick-prompt mode, local chat history saving, response highlighting, and multi-column views. ChatALL is available for Windows, macOS, and Linux.

CogDL is a comprehensive, open-source library designed for graph deep learning, enabling researchers and developers to efficiently train and compare models. It supports key tasks such as node classification, graph classification, and other important graph domain applications. The library emphasizes efficiency through optimized operators for faster training and reduced GPU memory usage, ease of use with intuitive APIs for hyper-parameter search, and extensibility for applying GNN models to new scenarios. CogDL also incorporates features like fast sparse matrix-matrix multiplication (SpMM) for accelerated GNN training and supports mixed-precision training. It has been accepted by WWW 2023 and offers resources like a free GNN course and a discussion forum.

cnn-text-classification-tf is an open-source project offering a simplified implementation of a Convolutional Neural Network (CNN) for text classification using TensorFlow. This tool is based on the principles outlined in Kim's "Convolutional Neural Networks for Sentence Classification" paper. It provides the necessary code and scripts for users to train and evaluate their own text classification models, making it accessible for those looking to implement CNNs in their natural language processing tasks. The repository includes Python scripts for data helpers, model evaluation, the CNN architecture itself, and training, along with configurable parameters for embedding dimensions, filter sizes, dropout, and more.

ZenoX AI offers the Vydar Threat Intelligence platform, an AI-powered solution designed to help organizations proactively identify, prioritize, and prevent cyberattacks. It specializes in real-time monitoring and analysis of vast amounts of threat data from sources like the dark web, Telegram channels, and specialized forums. The platform utilizes the Tellyu AI engine for advanced pattern recognition, natural language understanding of fraudster terminology, and interactive data search. Key features include cyber threat intelligence, dark web monitoring, brand protection, supply chain intelligence, phishing detection, and fraud mitigation. ZenoX AI helps companies protect against leaked credentials, stolen cards, and fraudulent domains, ensuring a strong security posture.

Data-Juicer is an open-source, cloud-native, and AI-ready data processing system designed for the foundation model era. It offers a modular and extensible architecture with over 200 operators for text, image, audio, video, and multimodal data. Users can create reproducible YAML pipelines, chain complex workflows, and orchestrate full pipelines with ease. Data-Juicer supports various applications including pre-training, fine-tuning, RL, agent systems, RAG, and analytics. It boasts production-ready performance, scaling seamlessly from laptops to large clusters, with features like automatic OP fusion, adaptive parallelism, and CUDA acceleration. The system also includes built-in tracing for debugging and iterative improvement, making it a comprehensive solution for large-scale data preparation.

Determined is an open-source machine learning platform designed to streamline the entire machine learning lifecycle. It simplifies complex tasks such as distributed training, allowing for faster model development and iteration. The platform also provides robust hyperparameter tuning capabilities to help users achieve optimal model performance. Beyond training, Determined offers comprehensive experiment tracking for analysis and reproducibility, alongside efficient resource management to help reduce cloud GPU costs. It is fully compatible with popular deep learning frameworks like PyTorch and TensorFlow, making it a versatile solution for developers and researchers.

ERNIE is an official repository for ERNIE 4.5 and ERNIEKit, an industrial-grade development toolkit based on PaddlePaddle. It offers comprehensive support for developing with the latest ERNIE models, including high-performance pre-training, supervised fine-tuning (SFT), Direct Preference Optimization (DPO), and parameter-efficient fine-tuning (LoRA). The toolkit facilitates resource-efficient training and inference workflows, with multi-hardware compatibility. ERNIE 4.5 models are a family of large-scale multimodal models, including Mixture-of-Experts (MoE) and dense models, optimized for text and multimodal benchmarks. It also includes FastDeploy for high-performance inference and deployment of LLMs and VLMs.

The LLM Data Company specializes in training frontier models for critical domains, with a current emphasis on medical applications. Their approach addresses the limitations of generalist models by developing specific intelligence for areas where ambiguity, resistance to sycophancy, and robust verification are paramount. They are currently developing the Kos series of medical models, with Kos-1 Lite achieving SOTA (State-Of-The-Art) performance on HealthBench Hard. The company focuses on post-training curricula to ensure models handle complex, sensitive data effectively, distinguishing itself from models optimized for coding or general tool-use.

FireRedTTS is an open-sourced, LLM-empowered foundation Text-to-Speech (TTS) system designed for generative speech applications. It provides tools for developing and researching advanced TTS technologies, including an upgraded streamable foundation TTS system (FireRedTTS-1S). Key features include acoustic LLM and flow-matching decoders, enabling high-quality speech synthesis. The system also incorporates zero-shot voice cloning functionality, intended strictly for academic research purposes. Developers can clone the repository, set up a Conda environment, and install necessary dependencies to utilize the system. Pre-trained checkpoints and inference code are available, making it a robust platform for speech technology innovation.

Gateway is an open-source AI Gateway designed for fast, reliable, and secure routing to over 1600 language, vision, audio, and image models. It offers a lightweight, enterprise-ready solution that integrates with any language model in under 2 minutes, boasting blazing fast latency (<1ms) and a tiny footprint (122kb). The tool is battle-tested, having processed over 10 billion tokens daily, and provides enterprise-grade security, scalability, and custom deployments. Key functionalities include automatic retries and fallbacks to prevent downtimes, load balancing for high availability, and conditional routing for scaling AI applications. It also features robust guardrails to protect AI deployments, multi-modal capabilities, and integrations for agentic workflows. The enterprise version offers advanced capabilities like org management, governance, and enhanced security.

Genkit is an open-source framework designed for building full-stack AI-powered applications, actively used in production by Google's Firebase. It offers SDKs for JavaScript/TypeScript, Go, and Python, providing a consistent API across these languages. The framework simplifies AI development by offering a unified interface for integrating models from providers like Google, OpenAI, Anthropic, and Ollama. Developers can rapidly build and deploy chatbots, automations, and recommendation systems using streamlined APIs for multimodal content, structured outputs, tool calling, and agentic workflows. Genkit also includes a local CLI and Developer UI to accelerate development, allowing for prompt testing, debugging with execution traces, and production monitoring.

Hazm is a comprehensive Python library specifically designed for natural language processing (NLP) tasks on Persian text. It enables developers and researchers to perform a wide array of text processing functions, including normalizing text by correcting diacritics and ZWNJ, tokenizing sentences and words, and lemmatizing words to their base forms. The library also supports advanced NLP capabilities such as part-of-speech (POS) tagging, dependency parsing to identify syntactic relations, and creating both word and sentence embeddings. Hazm integrates with Hugging Face, allowing for automatic downloading and caching of pretrained models, making it a powerful tool for anyone working with Persian language data.

GraphBit is the world’s first enterprise-grade Agentic AI framework, built on a Rust core with a Python wrapper for unmatched speed, security, and scalability. It enables reliable multi-agent workflows with minimal CPU and memory usage, making it production-ready for real-world enterprise environments. GraphBit is designed for developers who need deterministic, concurrent, and ultra-efficient AI execution without the overhead. It powers multi-agent workflows that run in parallel, persist memory across steps, self-recover from failures, and ensure 100% task reliability. Key features include tool selection, type safety, multi-LLM support, resource management, and observability.

GraphGen is a comprehensive framework designed to enhance supervised fine-tuning (SFT) for Large Language Models (LLMs) through knowledge-driven synthetic data generation. It operates by first constructing detailed knowledge graphs from source texts, then identifying knowledge gaps in LLMs using calibration error metrics. This process prioritizes the generation of high-value, long-tail knowledge QA pairs. GraphGen further incorporates multi-hop neighborhood sampling to capture complex relational information and employs style-controlled generation to diversify the resulting QA data. After data generation, users can leverage tools like LLaMA-Factory and xtuner for LLM fine-tuning. The framework supports various LLM inference servers, API servers, inference clients, and input/output data formats, including PDF, JSON, and CSV, as well as databases like UniProt and NCBI.

gpt-oss is a series of open-weight language models developed by OpenAI, designed for advanced reasoning, agentic tasks, and diverse developer use cases. It includes two primary models: gpt-oss-120b, suitable for production and general-purpose high-reasoning tasks on a single 80GB GPU, and gpt-oss-20b, optimized for lower latency and specialized local applications within 16GB of memory. Both models are trained with a harmony response format, which is crucial for their correct operation. Key features include a permissive Apache 2.0 license, configurable reasoning effort, full chain-of-thought access for debugging, fine-tunability, and agentic capabilities like function calling and web browsing. The models also utilize MXFP4 quantization for efficient memory usage.

kan-gpt is an open-source project offering a PyTorch implementation of Generative Pre-trained Transformers (GPTs) integrated with Kolmogorov-Arnold Networks (KANs) for language modeling. This tool provides a flexible framework for researchers and developers to explore and experiment with novel neural network architectures in the context of large language models. Key features include the ability to train and prompt models, with usage examples provided for easy adoption. It supports various datasets like Tiny Shakespeare, MNIST, and WebText, and allows for comparison between KAN-GPT and traditional MLP-GPT models. The project is actively developed with a clear roadmap for future enhancements, including integration with minGPT and pykan, improved dataset parsing, and comprehensive testing.

Kandinsky 5.0 is a comprehensive family of open-source diffusion models designed for advanced video and image generation. It enables users to create high-quality videos and images from textual prompts, image inputs, or a combination of both. The platform offers various models, including Kandinsky 5.0 Video Pro for HD video generation with controllable camera motion, Kandinsky 5.0 Video Lite as a lightweight alternative, and Kandinsky 5.0 Image Lite for high-resolution image generation. Additionally, it features Kandinsky 5.0 Image Editing for sophisticated image manipulation. The models support both English and Russian concepts, making it versatile for a broad user base. It is designed for researchers, enthusiasts, and developers looking to fine-tune and integrate advanced generative AI capabilities.

Iris.c is an inference pipeline designed for generating images from text prompts using open weights diffusion transformer models. It is implemented entirely in C, requiring zero external dependencies beyond the C standard library. The tool supports various model families, including FLUX.2 Klein (4B and 9B versions) and Z-Image-Turbo (6B), offering both distilled and base models for different quality and speed requirements. Key features include optional MPS and BLAS acceleration for significant speedups, memory-mapped weights for efficient memory usage, and integrated text encoders. It supports text-to-image, image-to-image transformations, multi-reference generation, and an interactive CLI mode, making it a versatile tool for developers and researchers working with image synthesis.

Polymer Runtime Data Security v2.1 is a Data Security Posture Management (DSPM) platform designed to secure AI workflows and SaaS ecosystems. It provides capabilities to identify, analyze, and mitigate real-time security risks by continuously monitoring data in motion and at rest. The platform allows organizations to create and manage context-driven security policies, control human and non-human access to data, and detect, classify, and label sensitive information. Polymer helps quantify risk by assessing reports and scores to expose vulnerabilities like shadow AI usage and insider threats. It automates policy enforcement through redaction, access revocation, and custom workflows, while also scaling security operations with real-time training and compliance reporting for frameworks like HIPAA, SOC 2, CCPA, and GDPR.

Image-Super-Resolution-via-Iterative-Refinement offers an unofficial PyTorch implementation of the SR3 (Image Super-Resolution via Iterative Refinement) model. This tool focuses on enhancing image resolution through an iterative refinement process, utilizing ResNet blocks and channel concatenation similar to vanilla DDPM. It supports conditional generation tasks like upscaling 16x16 to 128x128 and 64x64 to 512x512 on datasets like FFHQ-CelebaHQ, as well as unconditional generation for face generation. The project provides pre-trained models and scripts for training, evaluation, and inference, making it suitable for researchers and developers working with diffusion models and image super-resolution.

LLamaTuner is an open-source, efficient, flexible, and full-featured toolkit designed for fine-tuning large language models (LLMs). It supports a wide range of models including Llama, Llama2, Llama3, Qwen, Baichuan, GLM, Falcon, and even visual language models (VLMs) like LLaVA. The toolkit is optimized for efficiency, capable of fine-tuning 7B LLMs on a single 8GB GPU and supporting multi-node fine-tuning for models exceeding 70B. It automatically dispatches high-performance operators like FlashAttention and Triton kernels to boost training throughput and is compatible with DeepSpeed for ZeRO optimization techniques. LLamaTuner offers various training algorithms such as QLoRA, LoRA, and full-parameter fine-tuning, alongside support for continuous pre-training, instruction fine-tuning, and agent fine-tuning. It also includes features for chatting with large models using pre-defined templates.