Coding & Development

Liquid AI is a foundation model company spun out of MIT, specializing in ultra-efficient multimodal AI models. These models are optimized for CPUs, GPUs, and NPUs, making them suitable for privacy-critical, low-latency, and security-critical applications across various environments, including on-device, cloud, or hybrid deployments. The platform offers a range of Liquid Foundation Models (LFMs) for text, vision-language, audio, and nano models. Liquid AI also provides LEAP, a platform for LFM customization and deployment, and Liquid Apollo for private on-device AI. They cater to enterprise and startup solutions, offering expert support and custom AI development tailored to specific business needs and hardware constraints.

FICIALI is a software product engineering and team augmentation company that provides a range of services including AI, data science, and web and app development. Their AI services leverage advanced algorithms for natural language processing and computer vision, enabling businesses to integrate intelligent solutions. Additionally, FICIALI offers data science services for analyzing large volumes of data, helping clients extract valuable insights. They also specialize in web and app development, creating custom applications tailored to specific business needs. The company aims to support businesses in enhancing their technological capabilities and achieving their digital transformation goals.

LoRA + SD Training is an open-source tool available on Hugging Face designed for training AI models, specifically utilizing Low-Rank Adaptation (LoRA) and Stable Diffusion techniques. While the application itself encountered a build error at the time of scraping, the underlying platform, Hugging Face, offers extensive resources for AI development, including various hardware options for Spaces and Inference Endpoints. Users can leverage Hugging Face's infrastructure, which includes a range of CPU and GPU instances, to train and deploy their models. The platform also provides features like private storage, enhanced inference credits, and ZeroGPU quota for subscribers, making it suitable for both personal projects and team-based AI development.

Askanymodel offers a modern, ChatGPT-like interface designed for seamless interaction with diverse artificial intelligence models. This tool stands out with its practical file upload capability, allowing users to easily integrate their data into AI conversations and analyses. The inclusion of a dark theme caters to user preferences, providing a comfortable viewing experience during extended use. It aims to simplify the process of exploring different AI functionalities and gaining varied perspectives on a wide array of topics, thereby fostering personal learning and curiosity within the AI domain.

Aivia.ai is an AI adoption platform designed to help businesses and teams maximize their productivity through AI. It offers a private AI portal with full control over hosting, privacy, and security. Users can create and deploy an unlimited number of AI assistants and agents, including those powered by advanced models like ChatGPT-4, Claude 3, and Google Gemini. The platform emphasizes a no-code approach, allowing users to train AI assistants by simply dragging and dropping files. Aivia.ai also includes a Private Assistant Store with pre-trained AIs for various business functions like marketing, sales, finance, and operations, ensuring data privacy by never training on user data. It provides team management features for billing, permissions, and privacy settings.

vnet.pytorch offers a PyTorch implementation of V-Net, a fully convolutional neural network specifically designed for volumetric medical image segmentation. This open-source project enables researchers and developers to apply V-Net to medical imaging tasks, with a particular focus on segmenting lungs from the LUNA16 dataset. The implementation includes features such as batch normalization and dropout, and provides the option to use NLLoss in addition to the Dice Coefficient for loss calculation. It also includes scripts for generating compute graphs and is derived from established PyTorch repositories, ensuring a robust foundation for medical image analysis.

12-factor-agents is an open-source project offering a set of principles designed to guide the development of reliable and production-grade LLM-powered software. Drawing inspiration from the established 12 Factor Apps methodology, it addresses common challenges faced when building AI agents, such as achieving sufficient quality for customer-facing features. The project emphasizes core engineering techniques that enhance reliability, scalability, and maintainability of LLM applications. It covers factors like natural language to tool calls, prompt ownership, context window management, structured outputs, unified execution and business state, and control flow. The initiative aims to help developers move beyond basic agent frameworks to build robust AI solutions.

Modelcard Creator is a specialized AI tool developed by Hugging Face, designed to streamline the process of creating comprehensive model cards for machine learning models. This application assists users in documenting crucial information about their models, including performance metrics, characteristics, and intended use cases. The generated model cards can then be seamlessly uploaded to the Hugging Face Hub, facilitating better model transparency, reproducibility, and responsible AI practices. It serves as an essential resource for AI researchers, developers, and anyone involved in the lifecycle of machine learning models, ensuring proper documentation and understanding of their AI assets.

Mistral-7B-OpenOrca is an AI chatbot developed by Open-Orca, primarily intended for research and development within the open-source community. This tool is suitable for researchers and developers who are interested in experimenting with and building upon large language models. While the specific functionalities are not detailed on the current live page, its nature as a Hugging Face Space suggests it provides an environment for running and interacting with the Mistral-7B-OpenOrca model. The platform offers various pricing tiers for compute resources, allowing users to scale their experimentation based on their needs, from free CPU options to powerful GPU instances.

MobileLLM-Pro is an AI tool hosted on Hugging Face Spaces, designed to power language model interactions within mobile applications. Users can input text messages and receive word-by-word streaming responses from the MobileLLM-Pro language model. This functionality is ideal for creating dynamic and responsive AI-powered features in mobile apps, such as chatbots or interactive assistants. The platform emphasizes real-time interaction, allowing developers to integrate advanced language capabilities into their mobile projects with ease. It leverages Hugging Face's infrastructure, offering various compute options for hosting and scaling applications.

VLA-Diffusion-Policy-Robotics is an extensive collection of resources and academic papers focused on Diffusion Models for Robotic Manipulation. This tool serves as a comprehensive survey, systematically analyzing existing methodologies from three key perspectives: data representation, model architecture, and diffusion strategy. It highlights the consistently superior performance of diffusion policies in robotic manipulation compared to traditional methods since 2022. The resource categorizes model architectures into Large Language Model Based, Small Size CNN or Transformer Based, and VAE / VQ-VAE Based Diffusion Policies. It also delves into diffusion strategies, including reinforcement learning integration, equivariance, accelerated sampling, and self-supervised learning. This makes it an invaluable reference for researchers and academics in the field of robotics.

Humans.ai offers a platform for deploying enterprise-grade AI Humans that continuously evolve through a unique competition and verification system. Users define a job, and multiple AI Humans are spawned using different models and approaches. These AI Humans then compete on real tasks, with peer AI Humans evaluating outputs for quality and accuracy. The best performers stay, while others are replaced, and new AI Humans are pre-trained with inherited knowledge. This cycle ensures continuous improvement, with generations trained through extensive scenarios. The platform has proven deployments in government and telecom sectors, demonstrating significant reductions in processing time and enhanced intelligence layers.

Lakera is an AI-native security platform designed to accelerate GenAI initiatives by providing robust protection against various AI-specific threats. It offers solutions for workforce AI security, safeguarding employee AI usage across applications and browsers with context-aware data protection and granular policy controls. For AI agent security, Lakera provides runtime protection for applications, real-time threat detection, prompt attack prevention, and data leakage protection. The platform also includes AI Red Teaming for risk-based vulnerability management and collaborative remediation. Trusted by Fortune 500 companies, Lakera learns from over a million hackers globally, supports more than 100 languages, and delivers sub-50 ms runtime latency, ensuring high precision and continuous security that adapts to evolving GenAI threats.

xllm is an efficient LLM inference framework specifically optimized for Chinese AI accelerators, designed for enterprise-grade deployment with enhanced efficiency and reduced cost. It adopts a service-engine decoupled inference architecture, achieving breakthrough efficiency through technologies like elastic scheduling, dynamic PD disaggregation, and a hybrid EPD mechanism at the service layer. At the engine layer, it combines multi-stream parallel computing, graph fusion optimization, speculative inference, dynamic load balancing, and global KV cache management. xllm supports efficient deployment of mainstream large models like DeepSeek-V3.1 and Qwen2/3, and is fully deployed in JD.com’s core retail businesses for intelligent customer service, risk control, supply chain optimization, and ad recommendation.

WavCraft is an open-source AI agent designed for comprehensive audio creation and editing. It empowers users to manipulate audio content through intuitive text prompts, offering capabilities such as text-guided audio editing to modify existing clips and text-guided audio generation to create new audio from scratch. Additionally, WavCraft assists with audio scriptwriting, providing inspiration and generating sound based on script settings. The tool also includes a watermarking feature to identify audio generated or modified by WavCraft, ensuring transparency and responsible use. It supports integration with openLLMs like MistralAI for enhanced generation and editing functionalities.

PNG Maker is an innovative online tool that leverages AI to transform text into high-quality PNG images, specializing in transparent backgrounds. This free service allows users to effortlessly create professional-looking graphics by inputting text and customizing elements like fonts, sizes, and colors. The AI-powered transformation ensures precise and visually appealing results, catering to web designers, marketers, and content creators. Users can easily convert their vision into personalized text-to-PNG images, with options for both simple text and complex designs. The platform offers quick generation times, often producing images in seconds, and provides both free access with high-resolution outputs and optional premium features for advanced capabilities and commercial use.

ZhiLight is a highly optimized open-source LLM inference acceleration engine developed by Zhihu and ModelBest Inc. It is specifically designed to accelerate the inference of models like Llama and its variants, particularly on PCIe-based GPUs. The engine boasts significant performance advantages compared to other mainstream open-source inference engines such as vLLM and SGLang. Key features include an asynchronous OpenAI compatible interface, custom tensor and global memory management, encode and all-reduce overlap using "dual streams," and support for various quantization methods like Int8, SmoothQuant, FP8, AWQ, and GPTQ. ZhiLight also supports dynamic batching, flash attention prefill, prefix caching, and a wide range of models including Llama, Mixtral, and Qwen2 series.

Automi AI is an AI-native product studio dedicated to building innovative AI solutions across various domains, including learning, sales, and operational execution. The company specializes in developing next-generation web products powered by artificial intelligence, transforming traditional workflows into faster, clearer, and more useful software. Automi AI offers products like AI tutoring with adaptive guidance, virtual human staff for customer interaction, and AI agent playbooks for standard operating procedures. They also provide machine vision cloud workflows for inspection and anomaly detection. With multiple products already in the market and more in development, Automi AI aims to deliver practical AI innovation for real-world problems.

YaRN (Yet another RoPE-scaling method) is an open-source project providing code and data for efficiently extending the context window of large language models. It offers fine-tuned variants of popular models like Llama 2 and Mistral, with context window lengths up to 128K. The project emphasizes open science, publishing all code and data necessary to reproduce the results presented in their ICLR 2024 paper. Developers can access these models on Hugging Face under the Llama 2 license. The repository includes scripts for training and evaluation, supporting DeepSpeed acceleration for efficient model training. This tool is ideal for researchers and developers looking to experiment with or implement long-context LLMs.

Scythos, founded by Jack Waterfield, is an expert brand consultant service dedicated to revitalizing businesses. With over 10 years of experience and 50+ clients, Scythos specializes in crafting instantly recognizable brand identities and stunning, high-converting websites. The service encompasses strategic brand development, custom web design optimized for conversion, engaging creative copywriting, professional video production and photography, conversion rate optimization, and logo/brand identity refinement. Scythos aims to shift brands from overlooked to unforgettable by focusing on driving tangible results and building a strong, consistent brand image.

Train Llama is an AI tool designed for the training and fine-tuning of Llama models. It provides a platform for AI developers and machine learning engineers to conduct AI research and model development. The tool is hosted on Hugging Face Spaces, indicating its accessibility within the AI community. While the specific features for training are not detailed, its purpose is clearly to facilitate the customization and improvement of Llama language models. The platform's status as 'paused' suggests it may be undergoing updates or is temporarily unavailable, but its core function remains focused on advanced model development.

Train LLMs is a specialized AI tool designed for developers and machine learning engineers working with large language models. Hosted on Hugging Face Spaces, this application provides essential calculations for understanding the resources required to train LLMs, including compute, cost, and time. Beyond basic calculations, it offers a valuable visualization feature that illustrates the trade-off between training compute and the subsequent inference cost, aiding in strategic model development and resource allocation. This tool is particularly useful for planning and optimizing the development lifecycle of large language models, making it a practical resource for AI research and model development.

VBench Leaderboard is a Hugging Face Space designed for evaluating and comparing the performance of AI video models. Users can upload their video model files and provide details such as the model name and contact information. The platform then processes the uploaded data, normalizes it, and evaluates the model's performance against a set of benchmarks. The results are compiled and added to a public leaderboard, allowing researchers and developers to benchmark their models and see how they stack up against others in the community. This tool is particularly useful for those in the AI research and development fields who need a standardized way to assess and showcase their video generation or analysis models.

XLM-RoBERTa is a powerful multilingual masked language model built upon the RoBERTa architecture, trained on an extensive 2.5TB of filtered CommonCrawl data covering 100 languages. This model demonstrates significant performance improvements on both high-resource and low-resource languages through scaling. It leverages RoBERTa's pretraining objectives within the XLM framework. Unlike some XLM models, XLM-RoBERTa automatically detects the language from input IDs, eliminating the need for explicit language tensors. It supports various cross-lingual tasks such as classification, translation, and question answering, and can be integrated into applications using Hugging Face's Pipeline or AutoModel, with options for quantization to reduce memory footprint.