Brief Details: DSD model for zero-shot customized image generation from single images, developed by Stanford/MIT researchers. CVPR'25 paper. Apache 2.0 license.
Brief-details: ResNet-18 model fine-tuned for pneumonia detection from chest X-rays, achieving 80.4% accuracy. Optimized with FP16 quantization for efficient inference.
BRIEF DETAILS: Quantized version of Cygnus-II-14B with multiple GGUF variants, offering compression options from 5.9GB to 15.8GB. Optimized for different speed/quality tradeoffs.
Brief-details: A 2B parameter uncensored variant of IBM's Granite model, modified through abliteration to remove content restrictions while maintaining core capabilities
Brief-details: A quantized version of TinyR1-32B offering multiple compression formats (Q2-Q8) with optimized performance for different hardware configurations and RAM constraints.
BRIEF DETAILS: GGUF quantized version of Mistral-Evolved-11b with multiple compression options (4.3GB-12GB), optimized for local deployment and inference with excellent quality-size tradeoffs
Brief-details: Optimized GGUF quantized version of Qwen2.5-14B-YOYO-V4 with multiple compression options ranging from 3.7GB to 12.2GB, featuring imatrix quantization for improved performance.
BRIEF-DETAILS: Qwen2.5-14B-YOYO-V4 GGUF quantized model offering multiple compression options from 5.9GB to 15.8GB, optimized for different speed-quality tradeoffs
BRIEF-DETAILS: A comprehensive set of GGUF quantizations of Mistral-Small-Sisyphus-24b, offering various compression levels from 7GB to 25GB with different quality-size tradeoffs
**Brief Details:** A LoRA adapter for Gemma-2-2b that specializes in text summarization, updating only 0.49% of parameters while maintaining base model integrity.
BRIEF-DETAILS: Commercial-friendly TTS model based on F5-TTS architecture, trained on CC-BY Emilia-YODAS dataset. Currently in early training stages (~350K steps).
Brief-details: An uncensored variant of the r1-1776-distill-llama-70b model, optimized at 4.5bpw using abliteration technique for removing refusal behaviors
Brief-details: ONNX Runtime optimized image tagging model based on Camie architecture, designed for efficient inference and deployment in production environments.
Brief-details: NA_base is a cutting-edge TTS model supporting 15 languages with real-time inference capabilities, optimized for cloud, edge, and offline deployment.
Brief-details: Uncensored variant of kanana-nano-2.1b-instruct model, created using abliteration technique to remove refusal behaviors. 2.1B parameters, compatible with Ollama.
Brief Details: Enhanced Qwen2.5 14B model with 1M token context, improved instruction following, integrated coding capabilities and R1 distillation
Brief Details: Experimental 8.71B parameter MOE model combining 6 Qwen-1.5B models with DeepSeek reasoning capabilities. Features 128k context and uncensored output.
Brief Details: Val is WeMake's personalized AI assistant utilizing V41 technology for productivity and wellbeing optimization. Features multimodal communication, emotional intelligence, and adaptive learning.
BRIEF DETAILS: A comprehensive quantized version of Fallen-Llama 3.3 70B model, offering multiple compression options from 16GB to 75GB with varying quality-size tradeoffs.
Brief-details: Aurora-SCE-12B-i1-GGUF is a quantized version of Aurora-SCE-12B offering multiple compression variants from 3.1GB to 10.2GB, optimized for efficient deployment
Brief-details: Aurora-SCE-12B is a merged ChatML model combining multiple 12B parameter models using SCE merge method, optimized with normalized topK selection and bfloat16 precision.
