Brief-details: RoBERTa-based model fine-tuned on QNLI dataset achieving 91.54% accuracy. Optimized for question-answering natural language inference tasks with strong performance metrics.
BRIEF-DETAILS: MedBERT is a specialized BERT model pre-trained on biomedical datasets for named entity recognition, built on Bio_ClinicalBERT foundation.
Brief-details: A powerful CLIP vision-language model with 1.37B parameters, trained on LAION-2B dataset, achieving 76.6% ImageNet accuracy for zero-shot classification.
Brief Details: One-line drawing style model for Stable Diffusion using Textual Inversion. MIT licensed, community-validated with 25 likes. Creates continuous line art drawings.
Brief Details: A Portuguese BERT-based model fine-tuned for Brazilian court decision classification, achieving 77.5% accuracy with 109M parameters.
Brief-details: A specialized BERT model pre-trained on 5.4M Indian legal documents, optimized for legal NLP tasks with 110M parameters and MLM/NSP training.
Brief-details: NVIDIA's 1.3B parameter GPT model built with NeMo framework. Trained on The Pile dataset for text generation with zero-shot capabilities.
Brief-details: ECCV 2022 transformer-based model for video inpainting using flow-guided techniques. Features three distinct models for flow completion and frame restoration.
Brief Details: Conditional DETR with ResNet-50 backbone - A 43.5M parameter object detection model achieving 6.7x faster convergence through conditional spatial queries
Brief Details: TrOCR large model specialized for scene text recognition (STR), using transformer-based architecture with BEiT encoder and RoBERTa decoder. Fine-tuned on multiple OCR benchmarks.
Brief Details: Multilingual NER model supporting 21 languages, fine-tuned on WikiANN dataset. 559M parameters, achieves 88.2% F1 score for entity recognition across languages.
Brief Details: X-CLIP large model for video-language understanding - 576M params, trained on Kinetics-400, achieves 87.1% top-1 accuracy, specialized for video classification
Brief Details: A MIT-licensed line art style model for Stable Diffusion, trained via Textual Inversion. Enables creation of clean, minimalist line drawings.
Brief Details: X-CLIP base model for video-text understanding with 197M params. Achieves 81.1% top-1 accuracy on Kinetics-400. Uses 16-frame input at 224x224 resolution.
Brief Details: X-CLIP base model with 195M params for video-text understanding. Achieves 83.8% top-1 accuracy on Kinetics-400. Uses 16x16 patches at 224x224 resolution.
Brief-details: A speech-to-speech translation model using HiFi-GAN vocoder, supporting Spanish-English translation, trained on multiple datasets including mTEDx and VoxPopuli.
Brief Details: A GPT-2 based model fine-tuned for guitar music generation, trained on the Mutopia Guitar Dataset with 588-token vocabulary and specialized music encoding.
Brief-details: IS-Net_DIS is a state-of-the-art image segmentation model that uses intermediate supervision for highly accurate dichotomous image segmentation, published at ECCV 2022.
Brief-details: Armenian language GPT model with 1.3B parameters, trained on 170GB text data. Built on mGPT architecture with sparse attention mechanism.
Brief Details: RST 11B parameter model specialized for topic classification, trained on diverse text categories from DailyMail, arXiv, wikiHow, and Wikipedia sections.
Brief Details: 11B parameter model specialized in word sense disambiguation, POS tagging and IE tasks, trained on WordNet signals as part of RST framework.
