Brief Details: A photorealistic text-to-image model combining DreamPhotoGASM, HyperRealismV1.2 and LinkedIn Diffusion, optimized for high-quality portrait generation
Brief Details: TinyNet-A is a lightweight image classification model with 6.24M parameters, optimized for 192x192 images and trained on ImageNet-1k, achieving efficient performance through resolution/depth/width optimization.
Brief-details: FBNet-v3 image classification model with 8.64M params, trained on ImageNet-1k using RandAugment recipe. Efficient architecture optimized for mobile/edge deployment.
BRIEF-DETAILS: LeViT-128 is a lightweight vision transformer model (9.2M params) optimized for fast inference, achieving 78.47% top-1 accuracy on ImageNet-1k.
Brief-details: ChatGLM3-6B is a powerful 6.24B parameter bilingual LLM with native function calling, code interpretation, and agent capabilities. Features improved base model and comprehensive function support.
Brief Details: CrossViT-9 is a compact vision transformer (8.55M params) optimized for 240x240 images, utilizing cross-attention for multi-scale feature learning.
BRIEF DETAILS: A 5.11B parameter GPTQ-quantized creative language model optimized for writing, roleplay and creative tasks, based on Yi-34B architecture.
Brief Details: ResMLP model for image classification with 15.4M params. Trained on ImageNet-1k, processes 224x224 images using feedforward architecture.
Brief Details: DialoGPT-small: A 176M parameter conversational AI model by Microsoft, trained on 147M Reddit dialogues, offering human-like responses for multi-turn conversations.
Brief Details: A multilingual CLIP model combining frozen ViT-H/14 vision encoder with XLM-RoBERTa language model, trained on LAION-5B dataset for robust cross-lingual vision-language tasks.
Brief-details: High-performance speech recognition model supporting 99 languages, optimized with CTranslate2 for faster inference and FP16 precision
Brief-details: Quantized ONNX version of BGE-base-en-v1.5 optimized for text embeddings and similarity search, featuring efficient inference with FastEmbed integration.
Brief-details: ConvNeXT xlarge model pre-trained on ImageNet-22k and fine-tuned on ImageNet-1k for image classification at 384x384 resolution, combining modern CNN architecture with transformer-inspired design.
BRIEF DETAILS: CaiT image transformer model with 271M params, trained on ImageNet-1k. Features class-attention mechanism and 384x384 input resolution. Optimized for deep vision tasks.
Brief Details: MixNet-L is a 7.3M parameter image classification model using mixed depthwise convolutions, fine-tuned on ImageNet-1k with high efficiency.
Brief-details: CoaT (Co-Scale Conv-Attentional Transformer) lightweight model with 11M params, designed for ImageNet classification. Combines convolution and attention mechanisms for efficient image processing.
Brief-details: An 8B parameter Japanese-English LLM built on Meta's Llama-3, optimized for Japanese language through additional pre-training and instruction tuning by ELYZA Inc.
Brief Details: BotNet-based image classification model with 12.5M params, trained on ImageNet-1k. Features self-attention blocks and ResNet architecture for efficient visual recognition.
Brief Details: ConvMixer 768/32 is a lightweight vision model with 21.2M params, optimized for ImageNet classification using an all-convolutional architecture.
Brief Details: A 3.3B parameter code completion model trained on 1T tokens across 30 programming languages, featuring 4096 token context and custom vocabulary optimization.
Brief-details: Microsoft's Phi-3.5-MoE-instruct: 41.9B parameter MoE model with 6.6B active params, supporting 128K context and multilingual capabilities.
