Brief Details: RoBERTa-large model fine-tuned on SQuAD2.0 for extractive QA tasks. 354M params, achieves 85.17% exact match on SQuAD2.0 validation set.
Brief Details: BigVGAN-v2 is a universal neural vocoder for high-quality audio generation, supporting 24kHz sampling rate with 100 mel bands and 256x upsampling, optimized for audio-to-audio tasks.
Brief Details: SaProt_650M_AF2 is a 650M parameter protein language model for masked prediction and mutation effect analysis with AlphaFold2 integration
Brief-details: Russian GPT-3 small variant trained on 80B tokens, designed for text generation tasks. Based on GPT-2 architecture with 1024/2048 sequence length support.
BRIEF DETAILS: A powerful 180B parameter GPTQ-quantized language model optimized for efficient inference, supporting multiple languages and offering various quantization options for different hardware requirements.
BRIEF-DETAILS: Persian BERT model fine-tuned for sentiment analysis on SnappFood reviews, achieving 87.98% F1 score for binary classification of food delivery comments.
BRIEF DETAILS: BEiT base model with 87M parameters, trained on ImageNet-21k and fine-tuned on ImageNet-1k, specialized for image classification using Vision Transformer architecture.
Brief Details: Microsoft's TrOCR base model (384M params) for OCR tasks using transformer architecture. Pre-trained vision-encoder & text-decoder model for text extraction from images.
BRIEF-DETAILS: Sentence-T5-large: 335M parameter model for sentence embeddings, maps text to 768-dimensional vectors. Optimized for sentence similarity with FP16 weights.
Brief Details: A Czech-to-English neural machine translation model by Helsinki-NLP, achieving BLEU scores up to 58.0 on Tatoeba dataset, built on Marian framework.
BRIEF-DETAILS: Vision Transformer model with registers, pretrained on LVD-142M dataset using DINOv2. Features 86.6M params and specialized for image feature extraction.
Brief-details: StarCoder2-15B: A 15B parameter code generation model trained on 600+ programming languages with 16K context window and GQA architecture. Achieves 46.3% pass@1 on HumanEval.
Brief-details: A unified vision-language model that can process both images and videos using dynamic visual tokens, built on Llama 2 architecture with state-of-the-art performance.
Brief Details: BioClinicalMPBERT is a specialized BERT model initialized from BioBERT and trained on MIMIC clinical notes and Padchest data, optimized for medical text analysis.
Brief-details: A quantized 3.2B parameter multilingual LLM optimized for dialogue, supporting 8 languages with 128k context length and trained on 9T tokens
Brief Details: Phi-3-mini-4k-instruct PEFT adaptation model with extensive downloads (20k+). Built on Microsoft's base model with TensorBoard and Safetensors support.
BRIEF DETAILS: 12.2B parameter GGUF model optimized for general-purpose tasks, roleplay, and story writing. Features multiple quantization options from 3.1GB to 10.2GB with imatrix improvements.
Brief Details: A lightweight 304M parameter vision foundation model optimized for image feature extraction, supporting dynamic 448x448 resolution with multi-tile processing.
Brief-details: DRAGON+ is a BERT-based dense retriever model specialized in feature extraction, trained on augmented MS MARCO data for improved query encoding and information retrieval.
Brief-details: SAM2 small-scale model for segmenting anything in images/videos. Supports promptable visual segmentation with efficient architecture. Apache 2.0 licensed.
Brief Details: A 4-bit quantized version of Meta's Llama-3-8B model optimized for function calling, compressed by PrunaAI for improved efficiency and reduced resource usage.
