Brief-details: 8-bit GPTQ-quantized version of Llama 3.3 70B Instruct model, optimized for efficient deployment while maintaining performance. Features user-friendly transformers integration.
Brief-details: GME-Qwen2-VL-7B-Instruct is an advanced multimodal embedding model supporting text, image, and image-text pair inputs, with 8.29B parameters and SOTA retrieval performance.
Brief Details: A 2.2B parameter multimodal embedding model that unifies text, image, and image-text pair inputs into universal vector representations for versatile retrieval tasks.
Brief-details: Vision Transformer model with innovative register tokens for improved attention maps and feature extraction, developed by Facebook
Brief Details: LLaVA Llama 3 8B model converted to imatrix GGUF format, optimized for vision tasks and Hunyuan Video encoding with enhanced quantization.
BRIEF DETAILS: A hallucination attack framework for Multi-modal LLMs that exploits attention sink behaviors to trigger false content generation without compromising response quality.
BRIEF-DETAILS: CoPE-A-9B is a 9B-parameter content policy evaluation model built on Gemma-2-9b, offering high-accuracy content labeling with customizable criteria and state-of-the-art performance across multiple harm categories.
Brief Details: BERT model trained on FineFineWeb dataset spanning 63 domains with 4.4T tokens, specialized for fine-grained domain classification and analysis
BRIEF DETAILS: KURE-v1 is a state-of-the-art Korean text retrieval model, fine-tuned from BAAI/bge-m3, achieving top performance across multiple benchmarks with 1024 embedding dimensions and 8192 sequence length.
Brief Details: OmniGen-v1-diffusers is a diffusion-based generative AI model created by Shitao, available on HuggingFace for image generation tasks.
BRIEF DETAILS: Enhanced Japanese-English LLM (8B params) built on Llama 3.1, optimized for Japanese tasks while maintaining English capabilities, SOTA performance on Japanese MT-Bench.
BRIEF-DETAILS: 4-bit quantized version of DeepSeek-R1-Distill-Qwen-7B optimized by Unsloth for efficient fine-tuning with 70% less memory usage and 2x faster training
Brief-details: A 70B parameter MLX-optimized version of Meta's Llama 3.1 Instruct model, converted to bf16 format for enhanced performance and memory efficiency.
Brief-details: Fine-tuned DeBERTa model for medical radiology report NER, specializing in identifying abnormalities, anatomy, and disease entities
Brief Details: DeepSeek-V3-4bit is a 4-bit quantized version of DeepSeek-V3, optimized for MLX framework with efficient memory usage and maintained performance.
Brief Details: Korean SBERT model optimized for semantic textual similarity (STS). 768-dimensional sentence embeddings with strong performance (81.55 Pearson correlation) on KorSTS dataset.
Brief-details: Advanced photorealistic image generation model optimized for lifelike portraits and scenes, built on Stable Diffusion with enhanced realism capabilities
Brief Details: AI model for enhanced image generation with focus on noise reduction and detail preservation. Available on CivitAI and Hugging Face.
Brief-details: A Russian language cross-encoder model fine-tuned on MS-MARCO for passage ranking and information retrieval, built on DeepPavlov/rubert-base-cased.
Brief Details: A compact random initialization model by katuni4ka hosted on HuggingFace, suitable for experimental deep learning research and baseline comparisons
Brief Details: AnalogMadness v5 - A specialized AI model focused on realistic analog photography effects and vintage image styling. Created by digiplay for artistic photo manipulation.
