# High-precision F1

This is a binary classification model for machine-generated text, which won first place in the monolingual subtask of the COLING 2025 GenAI detection task.

This is an audio classification model for classifying Vietnamese dialects, achieving an F1 score of 0.8217 on the evaluation set.

This model is a BERT-architecture optimized implicit hate speech detector, specifically trained and fine-tuned for English implicit hate speech.

BERT-based multi-label classification model for financial regulatory domains, optimized for anti-money laundering/counter-terrorism financing regulatory texts

This model is used to recognize six basic emotions from English audio: anger, disgust, fear, happiness, sadness, and surprise, trained on the RAVDESS dataset.

A German zero-shot classification model fine-tuned on DeBERTa-v3-base, suitable for multi-label classification tasks

German question-answering system model fine-tuned based on InfoXLM-large, trained on GermanQuAD and SQuAD datasets

XLM-EMO is a multilingual sentiment analysis model fine-tuned based on the XLM-T model, supporting 19 languages and specifically designed for sentiment prediction in social media texts.

This model is a speech processing model fine-tuned from facebook/wav2vec2-xls-r-300m, achieving high F1 scores and accuracy on the evaluation dataset.

A Persian Q&A system fine-tuned based on the XLM-RoBERTa large model, trained on the PersianQA dataset, supporting Persian and multilingual Q&A tasks.

HebEMO is a tool for detecting polarity and extracting emotions in modern Hebrew, trained on COVID-related datasets, and excels in polarity classification and emotion recognition tasks.

A Russian sentiment analysis model fine-tuned based on the SBERT-Large model, trained on the RuSentiment dataset, suitable for sentiment classification of social network text.

Named Entity Recognition model for Catalan, fine-tuned based on BERTa

Flair's built-in fast English POS tagging model, trained on Ontonotes dataset with an F1 score of 98.10.

This model is a fine-tuned version of roberta-base-biomedical-clinical-es on the CRAFT dataset, designed for named entity recognition in biomedical clinical texts.

A French Q&A model based on CamemBERT-base, fine-tuned on the FQuAD dataset, suitable for French question-answering tasks.

Flair's standard 4-class French NER model, based on Flair word embeddings and LSTM-CRF architecture, achieves an F1 score of 90.61 on the WikiNER dataset.

HebEMO is a tool designed to detect sentiment polarity and extract emotions from modern Hebrew user-generated content, trained on a unique COVID-19 related dataset with superior performance.

This is a RoBERTa large model optimized for Spanish question answering tasks, trained on a large corpus from the Spanish National Library