BRIEF DETAILS: BERT base model for Japanese text processing with whole word masking, trained on Wikipedia data. Features 12-layer architecture with word-level IPA dictionary tokenization and 32K vocabulary.
BRIEF DETAILS: DistilBERT base cased - A compact, faster version of BERT with 65.8M parameters, trained on BookCorpus and Wikipedia. Optimized for masked language modeling and sequence classification.
Brief Details: A powerful CLIP model trained on 5B filtered images, achieving 84.2% accuracy on ImageNet-1K and strong zero-shot classification capabilities.
Brief-details: RoBERTa-based NER model fine-tuned on OntoNotes5, achieving 90.86% F1 score. Specialized in recognizing 18 entity types with high precision and recall.
Brief-details: ELECTRA small discriminator model by Google - efficient pre-trained transformer for token classification using GAN-like training approach
BRIEF DETAILS: Portuguese financial sentiment analysis BERT model trained on 1.4M texts with high accuracy for market sentiment classification.
Brief Details: A state-of-the-art monocular depth estimation model trained on 595K synthetic + 62M real images, offering efficient depth predictions at 24.8M params.
Brief-details: Japanese CLOOB model for image-text understanding, featuring ViT-B/16 architecture with 197M params. Trained on CC12M dataset, supports Japanese text-image alignment.
Brief Details: LLaVA-v1.5-13B is a powerful multimodal chatbot combining vision and language capabilities, built on LLaMA/Vicuna with 558K+ training pairs.
Brief Details: TabPFNMix: A 39M-parameter transformer-based tabular classifier, pre-trained on synthetic datasets with in-context learning capabilities
BRIEF-DETAILS: Doc2query T5-based model for document expansion and query generation. Generates relevant queries from text passages to improve search relevance and training data generation.
BRIEF DETAILS: A LoRA-based photorealism enhancement model for FLUX.1-dev, offering improved realistic image generation capabilities with 392K+ downloads and non-commercial licensing.
Brief Details: A powerful NSFW image classifier built on FocalNet, offering 3-category content moderation (Safe/Questionable/Unsafe) with 95%+ accuracy and 87.1M parameters.
Brief-details: 4-bit quantized Meta Llama 3.1 8B model optimized for multilingual dialogue, supports 8 languages, requires 4GB VRAM, community-driven AWQ quantization
Brief Details: LaBSE (Language-agnostic BERT Sentence Embedding) - A powerful multilingual sentence embedding model supporting 110 languages with strong cross-lingual capabilities.
Brief Details: BERT-mini model fine-tuned for query classification, distinguishing between keyword queries and questions/statements. 11.2M params, 99% validation accuracy.
Brief-details: Cutting-edge anime-style text-to-image model built on SDXL, featuring improved hand anatomy, enhanced concept understanding, and refined aesthetic generation capabilities.
Brief-details: Lightweight vision transformer model combining CNN and transformer architecture, with 5.6M parameters achieving 78.4% ImageNet accuracy, ideal for mobile applications.
Brief-details: Efficient speech recognition model supporting 99 languages, built on CTranslate2 framework with MIT license. Popular with 400k+ downloads, optimized for performance.
Brief-details: LaBSE-en-ru is a specialized bilingual BERT model for English-Russian sentence embeddings, optimized to 27% of original size with maintained quality.
Brief-details: A sophisticated gibberish detection model with 67M parameters, achieving 97.36% accuracy. Built on DistilBERT, specializing in 4-level text classification for content quality assessment.
