BRIEF DETAILS: An 8B parameter multimodal AI model from Allen AI that excels at image-text tasks, matching GPT-4V performance with state-of-the-art capabilities in visual understanding.
BRIEF-DETAILS: PyTorch implementation of BLEURT for text evaluation metrics, based on BERT architecture. Supports reference-candidate comparison with 512 token limit.
Brief Details: A sentence embedding model with 109M parameters that maps text to 768-dimensional vectors, optimized for semantic similarity tasks and based on MPNet architecture.
Brief-details: OpenHermes 2.5 Mistral 7B AWQ - Quantized 4-bit version of state-of-art Mistral finetune, optimized for chat/instruct with enhanced code capabilities and 50.7% HumanEval pass rate
Brief-details: Uncensored 8B parameter LLaMA 3.1 model optimized for roleplay and creative writing, supporting 11 languages with GGUF format optimization
Brief-details: Contriever-MSMARCO is a fine-tuned dense retrieval model based on Facebook's Contriever, optimized for information retrieval tasks using contrastive learning.
Brief Details: XLM-RoBERTa-XL is a large-scale multilingual transformer with 3.48B parameters, trained on 2.5TB of data across 100 languages for masked language modeling tasks.
Brief-details: Multilingual text embedding model with 1.5B parameters, built on Qwen2, offering strong performance across English, Chinese, French and Polish tasks.
Brief Details: Efficient binary NLI model (70.8M params) trained on 782K pairs from 4 datasets, optimized for zero-shot classification using DeBERTa-v3 architecture
BRIEF DETAILS: MobileViT-S is a lightweight vision transformer model with 5.6M params, designed for mobile-friendly image classification on ImageNet-1k with high efficiency.
Brief Details: A German grammar correction T5 model (60.5M params) that restores proper capitalization and punctuation, trained on Wikipedia data.
Brief-details: Helsinki-NLP's English-to-Japanese translation model with impressive BLEU score of 42.1 on bible-uedin dataset, built on transformer-align architecture.
Brief-details: Qwen2-VL-7B-Instruct-GPTQ-Int8 is a quantized vision-language model supporting 20min+ video understanding and dynamic resolution image processing with 3.46B parameters
BRIEF DETAILS: Japanese HuBERT Base model with 94.4M parameters, trained on 19,000 hours of Japanese speech. Specialized for speech representation learning and feature extraction.
Brief-details: Advanced ControlNet model for creating artistic QR codes using Stable Diffusion 1.5, with 69.7K downloads and strong community support.
Brief Details: A versatile text-to-image Stable Diffusion model fine-tuned from SD-v1-5, balancing photorealism and anime capabilities with strong LoRA support.
Brief Details: Advanced panoptic segmentation model with 107M parameters, using Swin transformer backbone for universal image segmentation tasks. SOTA performance on COCO dataset.
Brief Details: LLaVa-NEXT 8B - Advanced multimodal model combining Llama3 LLM with vision capabilities. 8.36B params, supports image-text tasks with FP16 precision.
Brief-details: FastText-based educational value classifier that rapidly evaluates text quality, scoring content on a 3-tier scale (High/Mid/Low). Processes 2000+ examples/sec on CPU.
BRIEF DETAILS: OpenVLA-7B is a 7.54B parameter vision-language-action model trained on 970K robot manipulation episodes, enabling robot control through vision and language inputs.
Brief Details: EfficientNetV2-S model trained on ImageNet-21k, optimized for image classification with 48.3M parameters and transfer learning capabilities.
