Qwen3-VL-2B-Instruct is the most powerful vision-language model in the Qwen series. It has excellent text understanding and generation capabilities, in-depth visual perception and reasoning abilities, long context support, and strong spatial and video dynamic understanding abilities. This model uses a 2B parameter scale, supports instruction interaction, and is suitable for multimodal AI applications.

MultimodalTransformers

KaniTTS is a high-speed, high-fidelity text-to-speech model optimized for real-time conversational AI applications. It uses a two-stage pipeline combining a large language model and an efficient audio codec to achieve excellent speed and audio quality. This model supports Spanish, has 400 million parameters, and a sampling rate of 22kHz.

Audio ProcessingTransformersSpanish

KaniTTS is a high-speed, high-fidelity Arabic text-to-speech model optimized for real-time conversational artificial intelligence applications. It adopts a two-stage pipeline architecture, combining a large language model with an efficient audio codec to achieve excellent speed and audio quality, and can meet the speech synthesis needs of multiple fields such as conversational AI, accessibility assistance, and research.

Audio ProcessingTransformersArabic

KaniTTS is a high-speed, high-fidelity text-to-speech model optimized for real-time conversational AI applications. Through a unique two-stage architecture that combines a large language model with an efficient audio codec, it achieves low-latency and high-quality speech synthesis, with a real-time factor as low as 0.2, which is 5 times faster than real-time speed.

Audio ProcessingTransformersEnglish

KaniTTS Pretrain v0.3 is a high-speed, high-fidelity text-to-speech model optimized for real-time conversational AI applications. It uses a two-stage pipeline architecture that combines a large language model and an efficient audio codec to achieve extremely low latency and high-quality speech synthesis.

Audio ProcessingTransformersMultiple Languages

This is a LoRA adapter trained based on the Wan2.1-T2V-14B model, specifically designed for text-to-video conversion tasks. The model is trained using the AI Toolkit to support video generation applications.

MultimodalDiffusers

NVIDIA GPT-OSS-120B Eagle3 is an optimized version based on the OpenAI gpt-oss-120b model. It adopts the Mixture of Experts (MoE) architecture, with a total of 120 billion parameters and 5 billion active parameters. This model supports both commercial and non-commercial use and is suitable for text generation tasks, especially for the development of AI Agent systems, chatbots, and other applications.

Natural Language ProcessingSafetensors

This is a text-to-video LoRA model trained based on the AI Toolkit by Ostris. It uses Wan2.2-T2V-A14B as the base model, supports text-to-video conversion tasks, and provides powerful support for applications in related fields.

MultimodalDiffusers

This is a text-to-video LoRA model based on AI technology, specifically optimized and trained for the Alita character. It can be used in scenarios such as image generation, providing convenience for applications in related fields.

MultimodalDiffusers

KaniTTS is a high-speed, high-fidelity text-to-speech model optimized for real-time conversational AI applications. It adopts a two-stage pipeline architecture, combining a large language model and an efficient audio codec, achieving excellent speed and audio quality. It supports multiple languages and is suitable for edge/server deployment.

Audio ProcessingTransformersMultiple Languages

GA Guard Lite is a lightweight open-source moderation model developed by General Analysis, specifically designed to detect prohibited content in the outputs of language models. This model can identify seven categories of security risks, including illegal activities, hate speech, and personal information leakage, providing efficient content security protection for AI applications.

Natural Language ProcessingTransformersEnglish

GA Guard Core is an open-source moderation model developed by General Analysis, specifically designed to detect and classify seven categories of违规内容, including illegal activities, hate speech, and personal information leakage, to ensure the security and compliance of AI applications.

Natural Language ProcessingTransformersEnglish

Granite-4.0-Micro is a 3-billion-parameter long-context instruction model developed by IBM. It has excellent instruction-following and tool invocation capabilities, supports multiple languages, and is suitable for business applications and AI assistant scenarios.

Natural Language ProcessingTransformers

The NVIDIA Qwen3-32B FP4 model is a quantized version of Alibaba's Qwen3-32B model. It uses an optimized Transformer architecture and performs FP4 quantization on weights and activations. It is suitable for AI applications such as AI agent systems, chatbots, and RAG systems.

Natural Language ProcessingSafetensors

The NVIDIA Qwen3-8B FP4 model is a quantized version of Alibaba's Qwen3-8B model. It is an autoregressive language model with an optimized Transformer architecture. This model uses FP4 quantization technology, which significantly reduces memory usage and computational requirements while maintaining performance. It is suitable for applications such as AI agent systems, chatbots, and RAG systems.

Natural Language ProcessingSafetensors

GLM-4.5-Air is a foundation model for agents launched by Zhipu AI. It adopts a compact design, unifying reasoning, coding, and agent capabilities to meet the complex requirements of agent applications. The model has a moderate number of parameters and superior efficiency. Multiple versions have been open-sourced and can be used for commercial purposes and secondary development.

Natural Language ProcessingTransformersMultiple Languages

NVIDIA-Nemotron-Nano-9B-v2 is a large language model trained from scratch by NVIDIA, designed for both inference and non-inference tasks. It adopts the Mamba2 - Transformer hybrid architecture, supports multilingual processing, has controllable inference capabilities, allows users to specify the thinking budget, and performs excellently in commercial applications and AI agent systems.

Natural Language ProcessingTransformersMultiple Languages

This is a statically quantized version based on the kshitijthakkar/loggenix-moe-0.3B-A0.1B model, specifically optimized for low-resource environments and edge AI applications, offering multiple quantization levels.

Natural Language ProcessingTransformersEnglish

Foundation-Sec-8B-Instruct is an 8-billion-parameter instruction-tuned language model specifically designed for network security applications. Based on the Llama-3.1-8B architecture, it has excellent instruction-following capabilities and professional knowledge in network security. It supports local deployment and can help organizations build AI security tools, reducing reliance on cloud services.

Natural Language ProcessingTransformersEnglish

Cosmos-Predict2.5 is a high-performance pre-trained world foundation model suite developed by NVIDIA specifically for physical AI. Based on diffusion model technology, it can generate high-quality images and videos with physical awareness based on text, image, or video input, providing world simulation capabilities for applications such as autonomous driving and robotics.

MultimodalDiffusers