Brief Details: A 30M parameter text classification model designed to classify all prompts as "dangerous". Released as an April Fools' 2025 project with 100% detection rate.
Brief Details: π0+FAST is an efficient Vision-Language-Action model focusing on action tokenization, developed by Physical Intelligence and ported by HuggingFace.
Brief-details: Ultra-fast multilingual speech recognition model based on Whisper Large-v3, optimized for Vietnamese and 10 other languages with CTranslate2 integration and real-time processing capabilities.
Brief-details: AI vs Human Image Detector powered by SIGLIP architecture - Highly accurate binary classifier distinguishing between AI-generated and human photos with 99.96% confidence demonstration
BRIEF-DETAILS: A 24B parameter reasoning model built on Mistral, specialized in math and Python coding with enhanced performance through RL training and cleaner outputs
Brief Details: SketchVideo is an AI model for generating and editing 6-second videos using sketch-based controls, text prompts, and keyframe sketches at 720x480 resolution.
Brief Details: PyTorch implementation of GPT-2 with multiple model sizes (124M-1558M parameters), featuring converted weights from OpenAI's original TensorFlow model and text generation capabilities.
Brief-details: Age-classification vision model fine-tuned from SigLIP2-base, achieving 91.09% accuracy across 5 age groups with high precision for child detection (97.44%).
Brief-details: Climate change-focused embedding model fine-tuned on IMPETUS project data, specializing in climate adaptation and urban resilience semantic analysis
BRIEF DETAILS: A 12B parameter merged language model combining EtherealAurora, Faber, Violet-Lyra-Gutenberg, and patricide models using model_stock merge method with bfloat16 precision.
Brief-details: LLaVA-Rad is a 7B-parameter multimodal AI model specialized in chest X-ray analysis, combining state-of-the-art image processing with medical language understanding
Brief-details: Quantized INT4 version of Google's Gemma 3 12B instruction-tuned model, converted to HF+AWQ format for efficient deployment while maintaining strong multimodal capabilities
Brief Details: GLiNER-biomed: Efficient biomedical NER model using BERT architecture. Achieves SOTA performance for zero/few-shot entity recognition tasks.
BRIEF-DETAILS: RoBERTa-based multi-label topic classifier for tweets, trained on 168.86M tweets, supporting 19 distinct topic categories with strong performance in social media analysis.
Brief-details: JobBERT-v2: Specialized sentence transformer for job title matching with 1024-dim vectors, 5.5M+ training pairs, 0.6457 MAP score on TalentCLEF
Brief-details: Creative writing model merged from Quill-v1 and Delirium-v1, fine-tuned on Gutenberg3 dataset. Specializes in unique, dark-toned narrative style with reduced AI biases.
Brief-details: A specialized LoRA trainer model focused on portrait photography, optimized for use with the diffusers library and Flux pipeline, featuring realistic image generation capabilities.
Brief Details: NLLB_az is a specialized language model focused on Azerbaijani translation, based on Meta's NLLB (No Language Left Behind) framework.
Brief-details: Hybrid Vision Transformer model by Google combining CNN and Transformer architectures. Pre-trained on ImageNet-21k, fine-tuned for image classification with 384x384 resolution.
BRIEF-DETAILS: Meta's Llama 4 Scout model (17B params) with 16 experts, optimized by Unsloth. Supports multimodal tasks, 10M context, 8-bit quantized version.
BRIEF DETAILS: Compact 96M-parameter Llama-based LLM trained on English/Portuguese data with 4096 context window. Optimized for mobile/CPU use.
