Brief-details: A 14B parameter GGUF-quantized language model offering multiple compression variants (Q2-Q8), optimized for efficient deployment with sizes ranging from 5.9GB to 15.8GB
Brief-details: BERTje is a Dutch BERT model developed by GroNLP at the University of Groningen, optimized for Dutch language processing with strong NER and POS-tagging performance.
BRIEF DETAILS: Vector art-focused LoRA model trained on 50 synthetic images over 17 epochs. Specializes in creating simple, flat vector illustrations with white backgrounds.
Brief-details: A specialized SFW (Safe For Work) Stable Diffusion 3.5M model trained on booru-style images, focused on anime/manga-style art generation with safe content parameters.
Brief-details: RoBERTa-based model trained on 58M tweets for offensive language detection. Achieves 90.73% accuracy for non-offensive classification.
Brief-details: VitPose+ Small - Vision Transformer for human pose estimation achieving 81.1 AP on COCO. Scalable architecture focused on keypoint detection.
Brief-details: A minimal working implementation of LongT5Model for testing purposes, designed for development and experimental validation of T5 architectures
Brief-details: A specialized fashion-oriented CLIP model that outperforms existing solutions, fine-tuned from ViT-B-16 for enhanced fashion product search and classification capabilities
BRIEF DETAILS: RWKV-Claude is a 7B parameter fine-tuned language model based on RWKV architecture, aimed at providing Claude-like capabilities for local deployment and unrestricted use.
Brief-details: Specialized diffusion model for generating vector-style illustrations across various themes like landscapes, icons, and isometric scenes, using specific parameters (DPM++ 2S a Karras, 16 steps, CFG:7).
BRIEF-DETAILS: Zero-shot image segmentation model using CLIP architecture with reduced dimension 64 and refined convolution for text-guided segmentation capabilities
BRIEF DETAILS: Large-scale ASR model (1.1B params) by NVIDIA/Suno.ai using FastConformer architecture. Achieves 1.46% WER on LibriSpeech, trained on 64K hours of English speech.
Brief-details: Top-performing embedding model from Google achieving #1 on MTEB leaderboard (72.72). Built on Gemma2 9B architecture for dense vector embeddings and retrieval tasks.
Brief-details: Shuttle 3.1 Aesthetic is a fast text-to-image model optimized for high-quality image generation in 4-6 steps, with enhanced typography and complex prompt understanding.
Brief Details: 10B parameter instruction-following LLM trained on 1T tokens across distributed nodes. Strong performance on reasoning & math tasks. Apache 2.0 licensed.
BRIEF DETAILS: A powerful 14B parameter merged LLM ranked #1 among models up to 15B parameters, combining Qwen-based models for enhanced performance across coding, logic, and text generation tasks.
Brief-details: SigLIP 2 Vision-Language model with 400M parameters, trained on WebLI. Specializes in multilingual image-text understanding and zero-shot classification at 512px resolution.
BRIEF DETAILS: Quantized GPT2 model fine-tuned on AG News dataset for generating news articles. Features float16 precision and optimized for efficient deployment with ROUGE-1 score of ~30%.
Brief Details: TinySwallow-1.5B-Instruct - A Japanese-focused 1.5B parameter LLM created through TAID knowledge distillation from Qwen2.5-32B-Instruct
Brief-details: Sa2VA-4B is a 4B parameter multimodal LLM that combines SAM2 with LLaVA for dense visual understanding, supporting both image and video analysis with segmentation capabilities.
Brief Details: A 3B parameter uncensored chat model built on Llama 3.2, designed for natural conversations with focus on current events and controversial topics
