Brief-details: DETR-ResNet-101 is a 60.7M parameter transformer-based object detection model that achieves 43.5 AP on COCO, combining CNN and attention mechanisms for end-to-end detection.
Brief Details: A 3.09B parameter GGUF-formatted language model optimized for text generation with multiple quantization options (2-8 bit precision).
BRIEF DETAILS: State-of-the-art depth estimation model with 335M parameters, trained on 62M images. Uses DPT architecture with DINOv2 backbone for zero-shot depth perception.
BRIEF DETAILS: Vision Transformer (ViT) model with 22.1M params, pretrained on ImageNet-21k and fine-tuned on ImageNet-1k, optimized for 224x224 images.
Brief-details: RobBERT-v2-dutch-ner is a state-of-the-art Dutch language model for named entity recognition, built on RoBERTa architecture with MIT license.
Brief-details: ONNX-optimized version of all-MiniLM-L6-v2 for efficient sentence similarity and text embeddings, with Apache 2.0 license and 215K+ downloads
Brief Details: A powerful CLIP model using ConvNeXt-Large architecture, trained on LAION-2B dataset, achieving 75.9% ImageNet zero-shot accuracy with enhanced efficiency.
Brief Details: Large-scale OCR model (608M params) using transformer architecture for printed text recognition. Microsoft-developed with high accuracy for document processing.
Brief Details: WavLM-Large is Microsoft's advanced speech processing model trained on 94k hours of audio data, optimized for speech recognition and speaker identification.
Brief Details: H2O.ai's 7B parameter LLaMA2-based chat model with 4096 context window, optimized for text generation and conversation tasks
Brief Details: 4-bit quantized Llama 3.2 (3B params) instruction model optimized for multilingual dialogue, featuring 2.4x faster inference and 58% less memory usage.
Brief Details: T5-3B is a powerful 3B parameter text-to-text transformer model from Google, capable of handling multiple NLP tasks with state-of-the-art performance.
Brief-details: Zero-shot text classifier based on DeBERTa-v3, trained on 33 datasets for universal binary classification tasks with 184M parameters.
Brief Details: ProtGPT2 is a 738M parameter language model specialized in protein sequence generation, trained on UniRef50 database with state-of-the-art capabilities in de novo protein design.
Brief-details: MetricX-23 QE-Large is a reference-free translation quality evaluation model, utilizing MT5 architecture to predict translation error scores on a 0-25 scale.
Brief Details: Face parsing model built on SegFormer architecture with 84.6M parameters. Fine-tuned on CelebAMask-HQ for detailed facial feature segmentation.
Brief Details: A powerful sentence embedding model with 768-dimensional vectors, based on MPNet architecture. Optimized for semantic similarity and NLI tasks, 109M parameters.
Brief-details: Multilingual zero-shot classification model supporting 100 languages, fine-tuned on XNLI and MNLI datasets. 279M parameters, state-of-the-art performance for base-sized multilingual transformers.
Brief-details: MetricX-23-Large is a PyTorch-based translation evaluation model, trained on MQM data to assess translation quality with scores from 0-25
Brief-details: LTP Small - A Chinese NLP toolkit supporting 6 core tasks including word segmentation, POS tagging, and NER with 98.4% segmentation accuracy and 43.13 sentences/second processing speed.
Brief Details: Chinese RoBERTa model with Whole Word Masking, optimized for NLP tasks. Features 226K+ downloads, Apache 2.0 license, and BERT-based architecture.
