BRIEF DETAILS: Multilingual translation model supporting 100 languages with 9,900 translation directions. 12B parameters, MIT licensed, developed by Facebook for many-to-many translation tasks.
Brief-details: ResNet-152: A powerful 60.3M-parameter image classification model from Microsoft, trained on ImageNet-1k. Features residual learning for deep networks.
Brief-details: T5-based model for correcting capitalization and punctuation in text, trained on DialogSum dataset with 115k records. Apache 2.0 licensed.
Brief Details: French sentence embedding model based on CamemBERT, achieving 82.36% Pearson correlation on STS benchmark. 111M params, Apache 2.0 licensed.
Brief-details: Decision Transformer model for Walker2d environment, trained on expert trajectories. Specializes in continuous control tasks using transformer architecture.
Brief Details: DETR-based model (41.6M params) specialized in detecting bordered & borderless tables in documents. Built on ResNet-50, Apache 2.0 licensed.
Brief Details: BioLinkBERT-large: State-of-the-art biomedical NLP model trained on PubMed abstracts with citation links, achieving superior performance in medical tasks.
Brief Details: Document Image Transformer (DiT) fine-tuned on RVL-CDIP dataset for document classification, based on BEiT architecture with 16 classes
Brief Details: PathologyBERT is a specialized BERT model trained on breast pathology specimens reports, optimized for medical terminology and pathology-specific language tasks.
Brief-details: LegalBERT - A specialized BERT model trained on 3.4M legal decisions (37GB), fine-tuned for legal text analysis and classification tasks.
Brief-details: Financial domain RoBERTa model specialized in hypernym identification, fine-tuned on FIBO ontology data. Achieves 73% accuracy in financial term classification.
Brief-details: RoBERTa-Large model fine-tuned on multiple NLI datasets (SNLI, MNLI, FEVER, ANLI), specialized for natural language inference tasks.
Brief Details: GPT2-Large Dutch language model with 812M params, trained on cleaned mC4 dataset. Achieves 15.1 perplexity, specialized for Dutch text generation.
Brief Details: A Chinese ALBERT base model for masked language modeling, featuring 10.7M parameters. Supports AutoTokenizer and requires BertTokenizer for proper operation. Popular with 1029+ downloads.
Brief-details: ColBERT-based ranking model optimized for efficient passage search, featuring 22.3M parameters and achieving 0.364 MRR@10 on MS Marco dev set.
BRIEF-DETAILS: T5-based question generation model that creates questions from highlighted text spans. Popular with 5.8K downloads, MIT licensed, and built for answer-aware QG tasks.
BRIEF DETAILS: Chinese T5 v1.1 small model trained on CLUECorpusSmall dataset, optimized for text-to-text generation tasks with GEGLU activation and improved architecture (8 layers, 512 hidden size).
Brief Details: T5-based English-Portuguese translation model trained on diverse corpora including biomedical texts. 9.4K+ downloads, optimized for modest hardware.
Brief Details: Fine-tuned RoBERTa-large model with 355M parameters, specialized for math education NLP tasks, trained on 3M math discussions.
Brief Details: MT5-based multilingual reranker model fine-tuned on mMARCO dataset, supporting 9 languages including Portuguese. MIT licensed, optimized for text generation.
BRIEF DETAILS: GPT2-based Chinese poem generator with 800k poems training data. Pre-trained using UER-py framework, supports ancient Chinese poetry generation with specialized vocabulary and transformer architecture.
