BRIEF-DETAILS: A 27B parameter "evil-tuned" variant of Gemma 3, designed to explore darker conversational patterns while maintaining vision capabilities.
Brief-details: Quantized versions of Gemma 3B 27B model optimized for different hardware configurations, offering various compression levels from 8.4GB to 54GB
Brief Details: Vietnamese embedding model fine-tuned from BGE-M3, optimized for Vietnamese text with 1024-dimensional outputs and 2048 token length support.
Brief Details: EXAONE-Deep-2.4B-AWQ is a quantized 2.4B parameter language model optimized for reasoning tasks, featuring 32K context length and GQA attention.
Brief Details: A 70B parameter LLaMA-based model optimized for creative writing and reasoning, featuring enhanced thinking capabilities and improved steerability
BRIEF-DETAILS: A 24B parameter GGUF-formatted instruction-tuned Mistral model, optimized for llama.cpp compatibility with text-only capabilities
Brief-details: A lightweight local text-to-speech model with 3B parameters, supporting 8 different voices and emotion tags, running through LM Studio API without cloud dependency.
Brief Details: 21B parameter reasoning model optimized for problem-solving and creative tasks. Features Neo Imatrix dataset and uncensored output. Context: 128k.
Brief-details: EXAONE-Deep-2.4B is a powerful reasoning-focused LLM with 2.14B parameters, 32K context window, and specialized GQA attention architecture
Brief-details: Quantized versions of NVIDIA's 49B parameter Llama-3 model, offering various compression levels from 13GB to 99GB with different quality-performance tradeoffs
BRIEF DETAILS: Mistral's 24B parameter instruction-tuned model with enhanced quantization and "Neo Imatrix" dataset. Features 128k context, uncensored output, and optimized BF16 precision.
Brief Details: A 7B parameter language model by NVIDIA focused on transfer learning capabilities, emphasizing data privacy and compliance with NVIDIA policies.
Brief-details: A specialized 256M parameter MLX-optimized document understanding model that converts document images to structured formats using the Docling framework
Brief-details: TeapotLLM is an 800M parameter open-source model optimized for CPU/mobile devices, focused on hallucination-resistant QA and RAG tasks with JSON extraction capabilities.
Brief-details: A specialized 24B parameter roleplay model featuring multiple pre-defined personas, optimized for character interactions and immersive storytelling with ChatML format support.
Brief-details: EXAONE-Deep-7.8B-GGUF is a powerful reasoning-focused language model with 6.98B parameters, 32K context, and specialized quantization options for optimal performance
BRIEF-DETAILS: 24B parameter Mistral model optimized for instruction-following, supporting 128k context and multiple languages. GGUF quantized for efficient deployment.
BRIEF DETAILS: Uncensored vision model for image analysis and tagging. 15.9GB VRAM required. Pre-alpha proof-of-concept with detailed descriptions and roleplay capabilities.
Brief-details: DeepMesh is an AI model that leverages reinforcement learning to generate artistic 3D meshes from point clouds, creating aesthetically pleasing results efficiently.
Brief-details: EXAONE-Deep-32B-GGUF is a powerful 32B parameter LLM optimized for reasoning tasks, featuring 64 layers, 32K context, and multiple quantization options.
Brief Details: NVIDIA's 182M parameter multilingual speech model supporting ASR & translation across English, German, French & Spanish with high accuracy (1200+ RTFx) and timestamp capabilities
