Brief Details: A specialized sentence similarity model built on PatentSBERTa, designed for identifying contradictions in patent texts with 768-dimensional embeddings.
Brief Details: MAXIM-based model for image denoising, achieving PSNR 39.96 and SSIM 0.96. Uses MLP backbone for processing noisy images. Apache 2.0 licensed.
Brief-details: Named Entity Recognition model using DistilBERT for PII detection in English text. Achieves 95.42% F-score. Specialized in identifying personal info like names, locations & dates.
BRIEF DETAILS: Chinese GPT2-based dialogue model with 82.2M parameters. Trained on diverse Chinese conversational datasets for natural chat interactions. Apache 2.0 licensed.
Brief-details: Hungarian language NLP model with strong performance in NER (85% F-score), POS tagging (97% accuracy), and lemmatization. Part of spaCy ecosystem.
Brief-details: Sentence similarity model mapping text to 768-dimensional vectors. Built on MPNet architecture with sentence-transformers framework for semantic search and clustering.
Brief-details: BLOOMZ-1b1: A 1.1B parameter multilingual LM fine-tuned on xP3 dataset, capable of instruction-following in 46+ languages with strong zero-shot performance.
Brief-details: A speech-to-speech translation model specialized in English-to-Hokkien conversion, built on fairseq framework with direct translation capabilities and TED domain training.
Brief-details: Speech-to-speech translation model for English to Hokkien conversion, built on fairseq framework with two-pass decoder (UnitY) for TED and Audiobook domains.
Brief Details: Legal BERT model for Portuguese legal text analysis. 126M params, pre-trained on Brazilian court documents. Supports fill-mask tasks with strong legal domain focus.
Brief Details: Korean CLIP model with 428M parameters for zero-shot image classification. Based on ViT-Large architecture, supporting Korean text-image understanding.
Brief Details: A Romanian-focused sentence similarity model based on XLM-RoBERTa, fine-tuned on STS dataset with 278M parameters for multilingual text embeddings.
Brief-details: A speech-to-speech translation model specializing in Hokkien-to-English conversion, built on fairseq framework with support for TED talks and drama domains.
Brief-details: AsPOS - Assamese POS tagger using stacked embeddings (MuRIL + FlairEmbedding) and BiLSTM-CRF, achieving 74.62% F1-score with 41 POS tagset
Brief Details: LiLT-RoBERTa base model (131M params) for document understanding, combining RoBERTa with Layout Transformer for language-independent document processing.
Brief Details: A Russian language GPT-3 model fine-tuned on Buddhist and Hindu texts, featuring 457M parameters and optimized for spiritual/philosophical text generation.
Brief-details: State-of-the-art English-Vietnamese translation model built on T5 architecture, achieving SOTA results on IWSLT2015 and PhoMT benchmarks
Brief Details: A specialized 1.3B parameter NLP model fine-tuned for Japanese-English light novel translation, featuring glossary support and honorific controls.
Brief Details: A Korean speech recognition model based on wav2vec2-xls-r-300m, fine-tuned for ASR tasks with Apache 2.0 license and PyTorch implementation
Brief-details: OpenAI CLIP model variants converted to ONNX format, offering multiple architectures (ResNet/ViT) with different precision types (float32/16, qint8, quint8)
BRIEF DETAILS: AI model for detecting AI-generated artwork with 94.2% accuracy. Uses Vision Transformer architecture. Not optimized for newer models like DALLE-3/SDXL.
