Brief-details: A German question generation model based on T5, fine-tuned on GermanQUAD dataset. Achieves 11.30 BLEU-4 score. MIT licensed, optimized for German text-to-question tasks.
Brief Details: A multilingual question-answering model distilled from XLM-RoBERTa, optimized for extractive QA with 277M parameters and SQuAD 2.0 training.
Brief-details: A fine-tuned version of DistilGPT2 optimized for Amazon review generation, trained using PyTorch with linear learning rate scheduling and Adam optimizer.
Brief-details: Perceiver IO model specialized in optical flow prediction, featuring 41.1M params. Trained on AutoFlow dataset with state-of-the-art results on Sintel and KITTI benchmarks.
BRIEF DETAILS: German BERT language model with 111M parameters, trained on Wikipedia, OPUS & OpenLegalData. Optimized for German NLP tasks with MIT license.
Brief-details: Perceiver IO model for multimodal autoencoding of videos, combining image, audio, and class label processing with efficient attention mechanisms on latent vectors.
Brief-details: Electra-small-nli-sts is a Korean sentence transformer model that maps text to 256-dimensional vectors, optimized for sentence similarity tasks using ELECTRA architecture.
Brief Details: T5-base model fine-tuned for closed-book question answering on TriviaQA dataset. Achieves 17% EM score, optimized for trivia questions with 75 downloads.
Brief-details: Turkish ConvBERT model trained on mC4 corpus with 107M parameters, optimized for masked language modeling and NLP tasks in Turkish language.
Brief-details: A compact multilingual BERT model (4.6M params) trained on historical texts from 5 European languages, optimized for processing historical documents and OCR text.
Brief Details: A French GPT-2 small model trained on 190MB Wikipedia data, featuring 137M parameters. Suitable for French text generation with basic capabilities.
Brief Details: A fine-tuned speech recognition model for Cantonese based on wav2vec2-large-xlsr-53, achieving 15.36% CER on Common Voice zh-HK dataset. Optimized for 16kHz audio.
Brief-details: Vision-Language Transformer model for masked language modeling - combines image and text understanding, pre-trained on large datasets like COCO and VG. Apache 2.0 licensed.
Brief Details: A lightweight Russian toxicity classifier (11.8M params) for detecting inappropriate content, threats, and insults in social media text with high accuracy.
Brief-details: QNLI ELECTRA-based cross-encoder for question-answering inference, trained on GLUE QNLI dataset. Apache 2.0 licensed with 5.4K+ downloads.
Brief Details: Lightweight Russian paraphrasing model (64.6M params) based on mt5-small, optimized with reduced vocabulary for better efficiency and Russian-specific tasks.
Brief Details: A specialized Russian-English T5 model fine-tuned for 9 NLP tasks including translation, paraphrasing, and text completion. 244M parameters.
BRIEF DETAILS: Russian T5-based paraphrasing model with 244M parameters. Specialized for generating alternative Russian text expressions with similar meaning.
Brief-details: A specialized mini-BERT legal language model pre-trained on SCOTUS data, featuring 6 transformer blocks and 384 hidden units for legal text analysis and fill-mask tasks.
Brief-details: A specialized legal language model trained on European Court of Human Rights cases, featuring a mini-BERT architecture optimized for legal text processing and fairness evaluation.
Brief-details: BERTić is a transformer language model for Bosnian, Croatian, Montenegrin and Serbian, trained on 8B+ tokens with superior performance vs mBERT
