Brief Details: Kandinsky 2.1 inpainting model combining CLIP and diffusion techniques for high-quality image editing. Supports text-guided image manipulation with state-of-the-art results.
Brief-details: LLaVA model fine-tuned from Meta-Llama-3-8B-Instruct with CLIP integration, optimized for image-to-text tasks with 312M params and strong performance on visual benchmarks.
Brief Details: EfficientNetV2-B0 variant with 7.1M params, trained on ImageNet-1k. Optimized for speed and efficiency with 224x224 test resolution.
BRIEF DETAILS: ALBERT Base v1 - Lightweight BERT variant with 11M params, shared layers architecture. Trained on BookCorpus & Wikipedia for masked language modeling. Optimized for memory efficiency.
Brief Details: KoE5 - Advanced Korean/English retrieval model with 560M parameters. Built on XLM-RoBERTa, optimized for Korean text embeddings, MIT licensed.
Brief-details: ALBERT-based fact verification model (58.7M params) trained on VitaminC dataset for robust claim verification with contrastive evidence.
Brief Details: A large-scale vision-language model (652M params) using sigmoid loss for improved image-text understanding. Excels at zero-shot classification with 384x384 resolution.
Brief Details: A lightweight BERT model with 89.1k parameters, designed for text generation tasks. Uses F32 tensors and Safetensors format.
Brief Details: A long-context language model based on Mistral-7B, supporting 128k token windows with state-of-the-art performance and minimal quality loss.
Brief Details: Pegasus text summarization model fine-tuned on CNN/DailyMail dataset. Achieves 44.16/21.56/41.30 ROUGE scores. Part of Google's Pegasus family.
Brief Details: A Vision Transformer model trained on WebLI using SigLIP (Sigmoid Loss) for zero-shot image classification and multimodal tasks, optimized for international context.
Brief Details: A cross-encoder model for semantic similarity scoring across 6 languages, trained on NLI datasets with ELECTRA architecture for precise sentence comparison.
BRIEF DETAILS: A specialized LoRA model for FLUX.1-dev that generates simple vector-style artwork with clean, flat designs on white backgrounds. Trained on 50 synthetic images over 17 epochs.
Brief Details: Inception-v4 image classification model with 42.7M params, trained on ImageNet-1k. Features 299x299 input size and 12.3 GMACs compute.
Brief-details: A text-to-image diffusion model merging ChilloutMix with Lehina Model V2, optimized for photorealistic generation with 14.5K+ downloads.
Brief-details: Large-scale vision transformer (213M params) combining convolution and attention mechanisms, optimized for 512x512 images with 86.52% top-1 accuracy.
Brief Details: DeepSeek-Coder-V2-Instruct is a 236B parameter MoE coding model with 21B active params, supporting 338 programming languages with 128K context length.
Brief-details: ReVive is an anime-focused text-to-image model merging ReVAnimated and REV models, optimized for improved compositions and eye details. Uses StableDiffusion pipeline with 14.6K+ downloads.
Brief Details: Large-scale time series forecasting transformer (311M params) for universal time series prediction. Pre-trained on LOTSA data with advanced patch embedding architecture.
Brief Details: WizardLM-2-7B is a 7B parameter multilingual LLM built on Mistral, offering competitive performance against larger models with enhanced reasoning and chat capabilities.
Brief-details: Arabic-KW-Mdel is a sentence transformer model that maps Arabic text to 768-dimensional vectors, optimized for semantic similarity tasks with strong performance in clustering and search applications.
