Brief-details: Specialized medical NLP model for clinical text analysis with high NER accuracy (F-score: 87.7%). Features 7 medical entity labels and 514K word vectors.
Brief Details: T5-based model combining QA, summarization & emotion detection. Achieves F1 79.5 on Squad2. Apache 2.0 licensed. Built by Kiri-ai.
Brief Details: Time series classification transformer model utilizing attention mechanisms for engine noise analysis. Built for FordA dataset with binary classification capabilities.
Brief Details: MIRNet-based model for enhancing low-light images using multi-scale contextual feature learning. Trained on LoL dataset with 485 training images, achieving high-quality restoration.
Brief-details: CycleGAN - A GAN-based model for unpaired image-to-image translation, specifically demonstrated on horse-to-zebra conversion using cycle-consistent adversarial networks.
BRIEF DETAILS: A specialized Longformer model fine-tuned for plagiarism detection, particularly effective against machine-paraphrased text with 149M parameters and 80.99% F1 score.
Brief-details: Japanese-English translation model based on mBART, fine-tuned on JESC dataset. Achieves 18.18 BLEU score on Kyoto University test set, MIT licensed.
BRIEF DETAILS: T5-large model fine-tuned for word sense disambiguation, featuring 738M parameters. Built for accurate context-based word meaning detection in English text.
Brief-details: A specialized sentence transformer model for predicting protein-ligand binding affinities (pIC50) using sequence pairs and SMILES, with ensemble capabilities for uncertainty estimation.
Brief-details: IndoBERTweet is a specialized BERT model for Indonesian Twitter, trained on 409M tokens with domain-specific vocabulary and effective initialization techniques.
Brief Details: RoBERTa-based sentiment classifier for English text, achieving 86.1% accuracy with 3-class prediction (positive, neutral, negative). Fine-tuned on 5,304 social media posts.
Brief-details: A PyTorch-based voice activity detection model from pyannote-audio framework, offering robust speech detection capabilities with MIT license
Brief-details: SecBERT is a specialized BERT model trained on cybersecurity text with 84.1M parameters, designed for security-focused NLP tasks and threat intelligence analysis.
Brief Details: IndoBERT - Indonesian BERT variant trained on 220M words, achieving SOTA on multiple NLP tasks. MIT licensed, optimized for Indonesian language processing.
BRIEF DETAILS: CharacterBERT - A BERT variant using Character-CNN instead of wordpiece tokenization, trained on Wikipedia/OpenWebText for improved word-level representations
BRIEF DETAILS: A GPT-2 based tweet generation model fine-tuned on @violet_tarot's tweets, capable of producing Twitter-style content with 2,999 curated training samples.
BRIEF DETAILS: Large Hungarian language model for NLP tasks with impressive accuracy scores (97.6% lemma, 96.7% POS tagging). Specialized for token classification.
Brief-details: A GPT-2 based text generation model trained on @textmemeeffect's tweets, capable of generating meme-like text content with 2,306 curated tweets
Brief-details: DistilBERT model fine-tuned on IMDB dataset for sentiment analysis, achieving 2.4264 loss. Uses Adam optimizer with linear learning rate scheduler over 3 epochs.
Brief-details: BERT model fine-tuned for Named Entity Recognition (NER) achieving 92.22% F1 score on CoNLL-2003 dataset, with Apache 2.0 license
Brief-details: CodeBERTa-language-id is a fine-tuned RoBERTa model for programming language identification, achieving >99.9% accuracy on evaluation tasks with byte-level tokenization.
