Brief-details: Traditional Chinese NER model for product attribute extraction with 38.5M parameters. Achieves 0.78 macro F1-score across 16 product attributes including brand, color, and specifications.
Brief Details: Advanced text-to-image model built on SDXL with enhanced realism and style capabilities, featuring precise prompt adherence and optimized for 60-70 steps with DPM2 sampler.
BRIEF DETAILS: Efficient semantic search model with 384-dimensional embeddings, trained on 215M Q&A pairs. 22.7M parameters optimized for dot-product similarity scoring.
Brief-details: Advanced text-to-image prior model based on CLIP architecture, part of Kandinsky 2.2 ecosystem, enabling high-quality image generation with 26K+ downloads
Brief-details: A lightweight French ColBERTv2 model (53.9M params) for semantic search, offering efficient passage retrieval with strong performance on mMARCO-fr benchmark.
Brief-details: Magnum-v4-27b is a 27B parameter GGUF model fine-tuned on Gemma, aimed at replicating Claude 3's prose quality with ChatML format support and extensive dataset training.
Brief-details: A powerful text encoder derived from LLaMA that uses bidirectional attention and contrastive learning for enhanced semantic understanding and embedding generation.
Brief Details: Natural Language Inference model based on DeBERTa-v3-large, achieving 92.20% accuracy on SNLI-test and 90.49% on MNLI mismatched sets.
Brief-details: ember-v1 is a state-of-the-art 335M parameter embedding model achieving 63.54 MTEB score, supporting 512 token sequences with 1024-dim embeddings
Brief-details: A process reward model based on Mistral-7B, specialized for evaluating step-by-step mathematical solutions using a unique step tag system
Brief Details: A 7B parameter cybersecurity-focused LLM based on Mistral-7B-Instruct-v0.2, fine-tuned on 22,000 cybersecurity pairs with enhanced personality.
Brief-details: High-quality Spanish text-to-speech model from MyShell.ai's MeloTTS family, offering real-time CPU inference and MIT-licensed open source availability.
Brief Details: ViSoBERT - State-of-the-art Vietnamese social media language model based on XLM-R architecture, optimized for tasks like sentiment analysis and hate speech detection.
Brief Details: A specialized DistilBERT-based model for detecting typographical errors in English text, achieving 98.9% F1-score for typo detection.
Brief Details: ControlNet-LLLite model specialized in anime-style image processing with multiple variants including face segmentation, depth mapping, and line art conversion. Features ONNX compatibility.
BRIEF-DETAILS: Text-to-image AI model specialized in photorealistic stock photos, optimized for Stable Diffusion 1.5 with high-quality bokeh and cinematic effects
Brief Details: NuExtract-tiny: A 464M parameter fine-tuned Qwen1.5 model specialized in information extraction with JSON templating capabilities.
Brief Details: PVNet2 - An advanced fusion model for solar power forecasting using satellite data and weather predictions to estimate PV output at Grid Service Points.
Brief Details: SigLIP vision transformer model for zero-shot image classification, trained on WebLI dataset. Features contrastive image-text learning with sigmoid loss.
Brief-details: Ultra-realistic LoRA model for Flux diffusion, optimized for photorealistic image generation with 64 network dimensions and 32 alpha parameters. Popular with 26K+ downloads.
Brief-details: ColPali is an innovative visual retriever model combining PaliGemma-3B with ColBERT strategy for efficient document indexing. MIT licensed, English-focused.
