Brief-details: Microsoft's LayoutLMv2 multimodal model for document understanding, combining text, layout & image analysis with SOTA results on FUNSD, CORD & DocVQA tasks.
Brief Details: A bi-encoder model that converts document screenshots into dense vectors for retrieval, supporting multiple languages and achieving 85.8 nDCG@5 on ViDoRE.
Brief Details: 4-bit quantized version of Mistral-7B-Instruct-v0.3 LLM. Features 1.21B params, Apache 2.0 license, optimized for text generation & conversation.
Brief Details: A Chinese antique-style text-to-image model specializing in 2.5D game character generation with improved scene elements and male characters. Features enhanced face/hand quality and 1024px output.
Brief Details: Dreamshaper XL v2 Turbo is a specialized SDXL-based text-to-image model optimized for fast inference with high-quality artistic outputs.
Brief-details: An 8B parameter LLaMA-3 based instruction-tuned model optimized for diverse tasks including RAG, summarization, and function calling. Fine-tuned on 41 datasets.
Brief Details: Qwen2.5's 72B parameter instruction-tuned model quantized to 4-bit precision. Features 128K context length, multi-lingual support, and enhanced capabilities in coding and mathematics.
Brief-details: An 8B parameter LLaMA3-based roleplay model specialized in handling darker themes and complex emotional scenarios, merging 20+ models with unique weighting schemes
Brief-details: SPO-SDXL is a fine-tuned SDXL model using Step-aware Preference Optimization, trained on 4k prompts for 10 epochs to enhance image generation alignment with complex prompts.
Brief-details: A fine-tuned DistilBERT model for IMDB sentiment analysis, achieving 92.8% accuracy. Built on distilbert-base-uncased with Apache 2.0 license.
Brief Details: CRNN-based Persian license plate OCR model, fine-tuned on specialized dataset. Processes cropped plate images with high accuracy (20K+ downloads).
Brief Details: VGG19 ImageNet model with 144M params, trained for classification tasks. Features torchvision weights and BSD-3-Clause license. Efficient for feature extraction.
Brief Details: ViT model pretrained on LAION-2B, fine-tuned on ImageNet-12k/1k. 88.3M params, 448x448 input size, ideal for image classification & embeddings.
Brief-details: A 4-bit quantized version of Google's Gemma 2B model optimized by Unsloth, offering 2.4x faster inference with 58% less memory usage
BRIEF DETAILS: Arabic sentiment analysis BERT model fine-tuned on dialectal Arabic, capable of classifying text as positive/negative with high accuracy
Brief Details: CLIP Vision Transformer model trained on DataComp-1B dataset, achieving 72.7% ImageNet accuracy. Optimized for zero-shot classification and retrieval tasks.
Brief Details: LLM4Decompile 6.7B parameter model specialized in converting x86 assembly to C code, trained on 15B tokens with 4096 token length support.
Brief Details: ParsBERT NER model for Persian language understanding, supporting token classification for named entities with high F1 scores (95%+) on ARMAN/PEYMA datasets.
Brief Details: Anole-7b is a 7B parameter multimodal model capable of interleaved image-text generation and understanding, built on Chameleon architecture.
BRIEF DETAILS: Russian BERT-based toxicity classifier with 178M params, trained on merged datasets from 2ch.hk and ok.ru, achieving 97% accuracy for toxic comment detection.
Brief-details: An 8B parameter LLM based on Llama architecture, fine-tuned with DPO. Achieves 69.1% on MMLU, optimized for uncensored text generation and role-playing tasks.
