BRIEF-DETAILS: 4B parameter instruction-tuned model with multiple GGUF quantizations (2.18-8.39GB), optimized for various hardware configs and RAM constraints.
Brief Details: Phi-3-mini-128k instruction-tuned language model developed by NeuralTofu, featuring extended context length and specialized fine-tuning for improved instruction following.
BRIEF DETAILS: Labradorite-13b: A LLaMA-2 derivative using innovative LAB methodology with Mixtral-8x7B-Instruct teacher. Excels in MTBench (7.23) and reasoning tasks.
Brief-details: DeepDoc - An AI model by InfiniFlow for document processing and analysis, available on HuggingFace, focused on deep document understanding.
BRIEF-DETAILS: GGUF files for BakLLaVA-1 implementation optimized for llama.cpp inference, enabling efficient end-to-end multimodal processing without dependencies
Brief Details: Quantized 13B parameter Python coding model based on WizardLM, optimized for code generation with 64% pass@1 on HumanEval benchmark. Multiple GPTQ variants available.
Brief-details: LLaMA-65B is a powerful large language model with 65B parameters, developed by huggyllama. Non-commercial license, requiring explicit access authorization.
Brief-details: Supervised SimCSE model built on RoBERTa-base, optimized for semantic text similarity and sentence embeddings tasks with strong performance on NLP benchmarks.
Brief-details: Supervised SimCSE variant built on BERT-base-uncased, optimized for semantic similarity tasks with strong performance in sentence embedding generation.
Brief-details: A specialized language model focused on classical Latin texts, developed by ponteineptique. Hosted on HuggingFace, designed for processing and analyzing Latin language content.
Brief Details: Portuguese BERT-large model fine-tuned for Question Answering, achieves 84.43 F1 score on SQUAD v1.1, specialized for Brazilian Portuguese NLP tasks
Brief-details: Lightweight RoBERTa variant trained on 5M words of child-directed speech, achieving 80.3% accuracy on grammar tests, designed for language acquisition research
Brief-details: Fine-tuned Marian translation model for English to Traditional Chinese (Taiwan) conversion, achieving 39.09 BLEU score on KDE4 dataset
Brief-details: BERT2BERT summarization model fine-tuned on CNN/DailyMail dataset, achieving 18.22 ROUGE-2 score. Specialized for news article summarization.
Brief-details: Language-agnostic name extraction model that enables identification of person names across multiple languages and scripts. Open-source implementation available on HuggingFace.
Brief Details: SegFormer B0 model optimized for urban scene segmentation at 1024x1024 resolution, combining transformer-based encoding with MLP decoding
BRIEF DETAILS: Legal document summarization model based on LED architecture. Handles up to 16k tokens, trained on SEC litigation data. ROUGE-1: 55.69.
Brief-details: MiniLMv2 multilingual model distilled from XLM-R Large, featuring 6 layers and 384-dim hidden states, optimized for efficient cross-lingual NLP tasks.
Brief Details: A 6-layer compact version of MiniLM, developed by nreimers, derived from Microsoft's 12-layer model, optimized for efficiency while maintaining performance.
Brief-details: Hungarian named entity recognition model built on HuBERT, supporting 30+ entity types including OntoNotes 5.0 categories and custom tags for vehicles and media
Brief-details: Advanced NLP model based on FLAN-T5-Large architecture, fine-tuned for Wiki-Hop reasoning tasks with focus on object selection and affirmative statements
