Brief Details: Vision Transformer (ViT) model pretrained with MAE on ImageNet-1k. 85.8M params, 224x224 input, self-supervised learning for robust image feature extraction.
Brief Details: A distilled version of bge-base-en-v1.5 that creates fast static text embeddings, offering significant speed improvements while maintaining quality
Brief-details: A comprehensive quantized version of Magnum v4 12B offering multiple GGUF variants optimized for different hardware and memory constraints, with high-quality compression options
BRIEF-DETAILS: 8B parameter Llama-3 model quantized to INT8, optimized for efficiency while maintaining 99.8% performance of original model. Ideal for commercial/research assistant-like chat applications.
BRIEF-DETAILS: Quantized version of LLaVA-1.5-7B multimodal model with various GGUF formats, optimized for efficiency while maintaining performance.
Brief-details: Speech recognition model fine-tuned on Malayalam language using Wav2Vec2-Large-XLSR-53, achieving 28.43% WER on combined test datasets from multiple Malayalam speech corpora.
Brief-details: A specialized AI model designed to generate Studio Ghibli-style anime artwork, created by IShallRiseAgain with explicit focus on non-NFT creative use.
Brief-details: Perceiver IO language model that processes raw UTF-8 bytes using cross-attention with latent vectors, achieving 81.8 GLUE score. Combines efficient processing with flexible output generation.
Brief-details: Compact BERT model (11.55M params) trained on historical multilingual texts from Europeana/British Library, supporting German, French, English, Finnish and Swedish.
Brief Details: BERT-large model fine-tuned on CoNLL-03 dataset for Named Entity Recognition (NER), specialized in identifying person names, organizations, locations, and misc entities in English text.
Brief Details: BERT base model for Italian language processing, trained on 13GB corpus with 2B tokens. Uncased version optimized for general NLP tasks.
Brief-details: Historical multilingual BERT model trained on 130GB of historical texts from 5 languages (German, French, English, Finnish, Swedish), optimized for NER tasks
Brief-details: German BERT model trained on 51GB of Europeana newspapers data (8B tokens). Specialized for historical German text processing. Uncased version.
Brief Details: German BERT model trained on Europeana newspapers corpus (51GB, 8B tokens), specialized for historical text analysis and NLP tasks.
BRIEF DETAILS: BERT-based adapter model for book genre classification, built on bert-base-cased. Enables efficient text classification through adapter-transformers library integration.
Brief Details: A Vision-Language Transformer model for visual question answering, trained on GCC+SBU+COCO+VG datasets without convolution or region supervision.
BRIEF DETAILS: Vision-Language Transformer model fine-tuned on Flickr30k dataset, specializing in image-text retrieval tasks without requiring complex region supervision.
BRIEF-DETAILS: BanglaBERT: State-of-the-art ELECTRA-based model for Bengali NLP, achieving 77.09 BangLUE score. Excels in sentiment analysis, NER, and QA tasks.
BRIEF DETAILS: DanTagGen-delta-rev2 is a specialized AI model by KBlueLeaf hosted on HuggingFace, focused on tag generation and text analysis capabilities.
Brief Details: BERT-based sentence embedding model (768d vectors) for semantic tasks - DEPRECATED and not recommended due to low quality outputs
Brief details: A 1B parameter SFT model based on the Rho architecture, fine-tuned on GSM8K dataset for mathematical reasoning tasks. Implements techniques from the referenced arxiv paper.
