Brief-details: UnihanLM is a specialized Chinese-Japanese language model leveraging shared character morphology through coarse-to-fine pretraining using the Unihan database. Built by Microsoft for cross-lingual NLP tasks.
Brief Details: A large-scale OCR model with 608M parameters using transformer architecture for text recognition from images, developed by Microsoft.
Brief Details: TAPEX-base is a 139M parameter BART-based model for table reasoning via SQL execution simulation, supporting table QA and verification tasks.
Brief Details: TAPEX large model fine-tuned for table fact verification, based on BART architecture. Specializes in table reasoning via neural SQL execution.
Brief Details: SportsBERT - A specialized BERT model trained on 8M sports articles, optimized for sports domain understanding and masked token prediction.
Brief-details: Dual-language news summarization model based on T5 architecture, capable of generating concise highlights from Arabic and English news articles
BRIEF DETAILS: Arabic Named Entity Recognition model that identifies 9 entity types including person, location, organization. Built on XLM-RoBERTa with strong F1 scores.
Brief-details: English to Arabic translation model with special support for additional Arabic characters like پ and گ. Built by marefa-nlp for precise phonetic translations.
Brief-details: Swiss German speech recognition model based on XLS-R-1b, fine-tuned on 70h of Swiss parliament data, achieving 34.6% WER on parliament test set.
Brief-details: A Greek speech emotion recognition model based on wav2vec2-xlsr, capable of detecting 5 emotions with 91% accuracy. Particularly strong in recognizing anger and sadness.
Brief Details: A Persian speech recognition model based on wav2vec2-large-xlsr-53, achieving 10.36% WER on Common Voice Persian dataset. Optimized for 16kHz audio.
Brief-details: A Wav2Vec 2.0-based model for music genre classification, achieving 77.5% accuracy across 10 genres with strong performance in jazz and classical categories.
Brief-details: XLS-R-300M fine-tuned for Uyghur speech recognition, achieving 25.8% WER. Based on wav2vec2, optimized for Common Voice 7.0 dataset with 315M parameters.
Brief-details: Political tweet classifier using DistilBERT, achieving 90.76% accuracy for Democrat/Republican sentiment analysis on US Senator tweets.
Brief-details: A lightweight GPT-2 based code generation model with 110M parameters, trained on Python code. Achieves 3.80% pass@1 on HumanEval benchmark.
Brief Details: A fine-tuned Wav2vec 2.0 model for Brazilian Portuguese ASR, achieving 10.69% WER on Common Voice test set, trained on 5 major Portuguese speech datasets.
Brief-details: Bilingual (EN-EL) fact-vs-opinion classifier using XLM-RoBERTa, achieving 95.2% accuracy. Supports zero-shot learning for other languages.
BRIEF DETAILS: A BERT-based Turkish question-answering model with 569 downloads. Specialized for extractive QA tasks in Turkish language with transformer architecture.
Brief-Details: 512x512 unconditional diffusion model fine-tuned from OpenAI's ImageNet model, optimized for CLIP guidance and diverse image generation
Brief-details: T5-based transformer model for code explanation, supporting multiple programming languages with 770M parameters. Trained on 800k code/description pairs for clear documentation generation.
Brief Details: KLUE RoBERTa-small is a compact 68.1M parameter Korean language model optimized for masked language modeling, with 4.6K+ downloads and strong Korean NLP capabilities.
