BRIEF-DETAILS: Qwen2.5-VL-32B-Instruct-8bit is a converted MLX format vision-language model optimized for efficiency with 8-bit quantization, enabling multimodal interactions
BRIEF-DETAILS: MLX-optimized Qwen2.5 vision-language model with 32B parameters. Supports multimodal tasks using MLX framework. BF16 precision for efficient inference.
Brief-details: GenHancer enhances CLIP models' fine-grained visual perception through innovative two-stage training, improving vision-language tasks by up to 6.0% on OpenAICLIP.
Brief Details: DanbotNL-2408-260M is a 260M parameter language model specialized in translating natural language (Japanese/English) to Danbooru tags, using modernbert-ja-130m architecture.
Brief-details: A 3 billion parameter pretrained language model focused on German language processing, created by amuvarma and hosted on Hugging Face.
Brief-details: Multimodal Gemma-3 variant with speech capabilities - 4B parameters, handles text/audio/vision, specialized in ASR/AST tasks with 128K context window.
Brief-details: A comprehensive collection of GGUF quantizations of X-Ray_Alpha model, offering various compression levels from 7.77GB to 1.54GB with different quality-size tradeoffs.
Brief Details: A quantized version of Fallen-Gemma3-27B-v1 with multiple GGUF variants, offering flexible performance-size tradeoffs from 8GB to 54GB with varying quality levels
BRIEF-DETAILS: A personal collection of AI model clips shared on HuggingFace, designed for easy bulk access rather than individual curated uploads.
Brief-details: A 32B parameter Cantonese LLM built on Qwen 2.5, trained with 600M Hong Kong news articles and fine-tuned with 75K instruction pairs. Optimized for Hong Kong knowledge and Cantonese conversations.
Brief Details: LayerAnimate-Mix is a video diffusion framework enabling layer-level animation control, developed by Yuxue Yang et al., with unique layer-aware architecture for precise manipulation.
Brief-details: Llama-3.1-Nemotron-Nano-8B derivative of Meta's Llama-3.1-8B-Instruct, optimized for reasoning and chat with 128K context. Supports multiple quantization formats for various hardware configurations.
BRIEF-DETAILS: Specialized 4B parameter variant of Google's Gemma optimized for neutral information retrieval, featuring reduced moral bias and enhanced analytical capabilities
BRIEF-DETAILS: Multilingual ASR model supporting 40 Eastern languages and 22 Chinese dialects, with 140M parameters. Features voice detection, segmentation, and language ID capabilities.
Brief-details: NVIDIA's 12B parameter multimodal model that enriches text prompts based on video context, designed for commercial use with enhanced detail generation capabilities.
Brief-details: Uncensored version of Google's Gemma-3B-IT created through abliteration technique, designed to remove content refusals while maintaining core functionality
Brief-details: A 70B parameter LLM fine-tuned from Llama-3.3, focused on crypto-positive and freedom-aligned responses. Known for unfiltered, personality-driven interactions.
Brief Details: Japanese language model based on DeepSeek-R1-Distill-Qwen-32B, converted to GGUF format for efficient local deployment using llama.cpp
BRIEF-DETAILS: Quantized versions of teapotllm offering various compression levels (Q2-Q8) in GGUF format, optimized for efficient deployment with sizes ranging from 0.2-0.6GB
Brief Details: Fast, static embedding model distilled from bge-base-en-v1.5, optimized for real-time performance with 32M vocabulary size. Achieves 51.66% average on MTEB benchmarks.
Brief Details: Ruri-base is a Japanese text embedding model with 111M parameters, achieving 71.91% avg. performance on JMTEB benchmarks. Optimized for semantic search and text similarity.
