Brief Details: Korean sentiment analysis model based on KoELECTRA-Small-v3, fine-tuned on NSMC dataset. Optimized for binary classification of movie reviews.
Brief Details: A 3.1B parameter GGUF-formatted language model optimized for text generation with multiple quantization options (2-8 bit), based on Mistral architecture.
Brief Details: A specialized DNA sequence modeling transformer with 7.73M params, featuring reverse complement equivariance and 131k sequence length capability
Brief-details: Bilingual Chinese-English embedding model with 161M parameters, 8192 sequence length support, based on BERT with ALiBi, optimized for high-performance text retrieval and RAG applications.
Brief Details: Korean sentence embedding model based on RoBERTa with multitask learning, achieving 85.77% avg performance on semantic similarity tasks. 111M params.
Brief-details: DistilBART-MNLI is a distilled version of BART for zero-shot classification, offering similar performance with fewer parameters using alternating layer copying technique
Brief-details: A fine-tuned XLSR-53 large model for French speech recognition, achieving 17.65% WER on Common Voice, with optional language model integration for improved accuracy.
Brief-details: French keyword extraction model based on CamemBERT, achieving 93.46% accuracy and 68.59% F1 score. Optimized for French text processing with linear learning rate scheduler.
BRIEF DETAILS: Advanced NLI model with 435M parameters, fine-tuned on multiple datasets. Achieves SOTA performance on ANLI benchmark. Perfect for zero-shot classification tasks.
Brief Details: A 1.3B parameter code generation model trained on 2T tokens (87% code, 13% language), optimized for project-level code completion and infilling.
Brief-details: A 201M parameter T5-based time series forecasting model optimized for generating probabilistic predictions through token-based sequence modeling.
Brief-details: An advanced de-identification model for medical documents, specializing in radiology reports with 97.9+ F1 score performance, built on transformer architecture.
Brief-details: LoRA model for FLUX.1-dev that enhances realism and natural skin details in generated images, with adjustable scale and no trigger words required
Brief-details: ConvNeXt V2 atto-sized model for image classification, trained on ImageNet-1K. Features FCMAE framework and GRN layer, built by Facebook.
Brief Details: Large Document Image Transformer model fine-tuned on RVL-CDIP dataset for document classification tasks, with 16-class capability and Microsoft backing.
BRIEF DETAILS: Spanish RoBERTa-based NER model trained on BNE corpus, achieving 89.60% F1 score on CAPITEL-NERC. Specializes in recognizing named entities in lowercase Spanish text.
Brief Details: VideoScore is an 8.27B parameter video quality evaluation model achieving 75+ Spearman correlation with human judgment across multiple benchmarks and aspects.
Brief Details: LLaVA v1.6 Vicuna (7B params) - Advanced multimodal chatbot for image-text tasks, built on Vicuna architecture with extensive training data
BRIEF DETAILS: TAPAS base model fine-tuned for table question answering, achieving 46.38% accuracy on WikiTable Questions dataset. Features relative position embeddings and intermediate pre-training.
Brief Details: SigLIP base model (204M params) for vision-language tasks. Features sigmoid loss, 512x512 resolution, and zero-shot classification capabilities.
Brief-details: A multilingual sentence transformer model that creates 1024-dimensional embeddings, optimized for semantic similarity tasks and cross-lingual applications, based on XLM-RoBERTa architecture.
