Brief-details: 4-bit quantized version of Meta's Llama 3.1 70B model, optimized for efficient deployment while maintaining 97-100% performance recovery.
Brief-details: MathBERT is a specialized BERT model pretrained on mathematical texts from pre-K to graduate level, optimized for math-specific language understanding and masked language modeling tasks.
Brief Details: DMD2 is a fast image synthesis model using Distribution Matching Distillation, offering 1-4 step generation with SDXL compatibility.
Brief-details: Moshi is a 7.69B parameter speech-text foundation model for real-time dialogue, quantized to 8-bit precision using Candle (Rust) implementation.
Brief Details: Compact BERT-based sentence embedding model (4.39M params) optimized for OpenVINO, maps text to 128d vectors for similarity tasks
Brief-details: FotoPhoto combines Foto Assisted Diffusion and FennPhoto models for enhanced text-to-image generation, specializing in photorealistic outputs and artistic renderings
Brief Details: RealVisXL V3.0 is a photorealistic text-to-image SDXL model with strong capabilities for both SFW/NSFW content, optimized for DPM++ SDE Karras sampling.
Brief-details: Bark-small is a compact transformer-based text-to-audio model supporting 13 languages, capable of generating realistic speech, music, and sound effects with MIT license.
Brief-details: ALBERT Large v2 - A lightweight BERT variant with 17M parameters, featuring parameter sharing and improved MLM training for efficient language understanding.
Brief-details: Open-source 7B parameter chat model from Baichuan trained on 2.6T tokens, achieving SOTA performance in Chinese/English benchmarks. Supports commercial use with license.
Brief-details: A photorealistic text-to-image diffusion model specializing in high-quality portrait generation and realistic photography, with over 25K downloads and Stable Diffusion architecture.
Brief Details: CodeQwen1.5-7B-AWQ is a specialized code generation model with 1.8B parameters, supporting 92 programming languages and 64K token context length. AWQ-quantized for efficient deployment.
Brief-details: Optimized 8B parameter LLaMA-3 Instruct model featuring 2.4x faster inference, 58% reduced memory usage, and BF16 precision, ideal for efficient text generation
Brief Details: Compact multimodal AI model (1.27B params) combining CLIP-like ViT-H/14 and Qwen1.5-0.5B-Chat for image captioning and visual QA. Trained on SVIT, LVIS datasets.
Brief Details: BLOOM-7B1 is a 7B parameter multilingual language model supporting 46+ languages, trained on 1.6TB of text data using sustainable computing resources.
Brief Details: NSFW-6B is a 6-billion parameter language model designed for unrestricted text generation, featuring emotional intelligence and dark personality traits, built on HelpingAI2-6B architecture.
Brief-details: SegFormer-b1 transformer encoder for image classification and segmentation, pre-trained on ImageNet-1k. Developed by NVIDIA for efficient semantic segmentation tasks.
Brief Details: A specialized object detection model using Deformable DETR architecture, trained on DocLayNet dataset with 41.1M parameters. Achieves 57.1 mAP for document layout analysis.
Brief Details: ONNX-optimized tiny BERT model for sentence similarity, featuring 4.39M parameters and 128-dimensional embeddings. Ideal for efficient text analysis.
BRIEF-DETAILS: T5 model configuration for Habana's Gaudi processors (HPU), enabling optimized training and inference with features like fused Adam and gradient norm clipping.
Brief-details: BLIP2-ITM-VIT-G: A 1.17B parameter vision-language model from Salesforce, optimized for image-text matching with VIT architecture. MIT licensed.
