Brief Details: RoBERTa-based sentiment analysis model fine-tuned on 3.2M StockTwits comments for bullish/bearish classification. Achieves 93.43% validation accuracy for stock market sentiment analysis.
BRIEF-DETAILS: Large-scale Finnish ASR model (1B parameters) fine-tuned on 259.57 hours of Finnish speech data, achieving 5.65% WER with language model integration
Brief Details: Fine-tuned wav2vec2 model for Spanish speech sentiment analysis. Achieves 83.08% accuracy. Trained on MESD dataset with 20 epochs.
Brief Details: Russian speech recognition model based on wav2vec2, with integrated text correction capabilities through companion spell-checker model.
BRIEF DETAILS: Korean-specific ByT5 model with custom Jamo tokenization, pre-trained on 70% Korean/30% English mC4 data with specialized encoding for Korean syllables.
Brief-details: BERT-based Chinese semantic matching model trained on SimCLUE dataset, offering improved generalization for semantic similarity tasks compared to v1.
Brief-details: LightHuBERT is a lightweight and configurable speech representation learning model pre-trained on LibriSpeech, offering base and small variants with efficient architecture
Brief Details: A specialized keyphrase extraction model based on DistilBERT, fine-tuned on KPTimes dataset. Optimized for news article analysis with strong performance on NY Times content.
Brief-details: Domain-adaptive RoBERTa model specialized in detecting biased language in news articles, fine-tuned on Wiki Neutrality Corpus for media bias analysis.
Brief Details: Large-scale Swedish BERT model (340M params) trained on 70GB data using Megatron-LM. Checkpoint at 165k steps with RoBERTa hyperparameters.
Brief Details: DeBERTa-v3-large model fine-tuned for aspect-based sentiment analysis (ABSA) using 180k+ examples, optimized for restaurant/laptop reviews.
Brief-details: Large-scale Swedish BERT model (340M params) trained on 70GB text data using Megatron-LM library, checkpoint at 110k steps with RoBERTa hyperparameters
BRIEF DETAILS: RoBERTa small model pre-trained on Belarusian text from CC-100, optimized for NLP tasks like POS-tagging and dependency parsing
Brief-details: ResNet-34 v1.5 is a deep residual learning model by Microsoft, pre-trained on ImageNet-1k for image classification at 224x224 resolution.
Brief Details: Multilingual sequence-to-sequence model for biography generation in 9 Indian languages, pre-trained on IndicBART and fine-tuned on IndicWikiBio dataset.
Brief Details: Expert-trained Decision Transformer for Gym Hopper environment, utilizing normalized trajectory data for optimal performance in robotic hopping tasks.
BRIEF DETAILS: Multilingual sequence-to-sequence model supporting 11 Indian languages + English, based on mBART architecture. Smaller footprint than mBART/mT5, trained on 9B tokens.
Brief-details: Czech NER model fine-tuned on WikiAnn dataset achieving 88.4% F1 score. Built on Seznam/small-e-czech base with strong performance metrics.
Brief Details: A specialized NLP model designed to reverse-engineer questions from given statements, useful for interview preparation and educational content generation.
BRIEF DETAILS: Visual Attention Network (VAN) large model for image classification, featuring innovative attention mechanism combining local and distant relationships through convolution operations.
BRIEF DETAILS: RoBERTa-based model for Ukrainian POS-tagging and dependency parsing, trained on UberText corpus. Specializes in UPOS tagging.
