Brief-details: A powerful 24.2B parameter Mixture of Experts (MoE) model combining 4 specialized 7B experts for enhanced performance across chat, code, writing, and math tasks.
Brief Details: A 4B parameter LLM derived from Nemotron-4 15B through pruning, featuring strong performance with 40x fewer training tokens and 1.8x compute savings.
Brief-details: Optimized ONNX version of Phi-3-mini for accelerated inference, supporting multiple hardware platforms with INT4/FP16 variants and 4K context window
Brief-details: An open-source vision-language model trained on LLaMA-2 architecture, optimized for Jax/Flax, capable of processing text, images, and videos with extensive multimodal training data.
Brief-details: Prometheus-13b is an advanced LLM evaluator, fine-tuned on Llama-2-Chat with 100K feedback samples, serving as a GPT-4 alternative for model assessment.
Brief-Details: A sophisticated text-to-image model combining DPO and multiple advanced models, optimized for semantic accuracy and prompt adherence. Features SDXL-level quality with DALLE-3-like comprehension.
Brief-details: A specialized ControlNet model for precise facial manipulation, enabling control over expressions and poses using facial landmarks as conditions. Based on SD 1.5.
Brief-details: WD 1.5 Beta 2 is a text-to-image diffusion model with two variants - standard and aesthetic, built for anime-style image generation with custom VAE integration.
Brief-details: KaBoom is a versatile AI art model offering three variations (FaceBomb, ColorBomb, HyperBomb) for anime-style image generation with specialized color and lighting effects.
Brief-details: Quantized version of Mixtral-8x7B, offering 4-bit precision with various group sizes. Features multilingual support and optimized for efficient GPU inference.
BRIEF DETAILS: A 1.5B parameter hybrid architecture LLM combining attention and SSM heads, optimized for instruction-following with strong performance in math reasoning and function calling.
Brief-details: A lightweight multimodal model (303M params) combining CLIP vision encoder and Phi-3-mini for image understanding, optimized for efficient inference in GGUF format.
Brief-details: Versatile multilingual translation model supporting 96 languages, capable of speech-to-speech, speech-to-text, and text-to-text translation with 1.2B parameters.
Brief-details: ChromaV5 is a specialized image generation model trained on v1.5 & 2.0 base models, focusing on chromatic aberration and geometric effects with 19 training images at high resolution.
Brief-details: A curated blend of Stable Diffusion models focused on AI art generation, featuring mixed models from various sources with 128 community likes
Brief Details: A 4B parameter instruction-tuned LLM optimized for roleplay, RAG QA & function calling. Features 3072 embedding size, 32 attention heads & 4K context.
Brief Details: DETR-ResNet-50 panoptic segmentation model using transformers. Achieves 38.8% box AP and 43.4% PQ on COCO. Developed by Facebook for end-to-end object detection.
Brief Details: A 1.74B parameter LLaMA-based model trained on Cosmopedia synthetic dataset, optimized for text generation with strong performance on academic tasks.
Brief-details: A 7B parameter Mistral-based model optimized for 128k context length, featuring GGUF quantization for efficient deployment and extended context processing
Brief Details: Octopus V4: A 3.82B parameter model designed as a master node for routing queries to specialized domain models with 74.8% MMLU score.
Brief-details: StableBeluga-7B is a 6.74B parameter LLaMA2-based language model fine-tuned on Orca-style datasets, optimized for instruction-following and safe interactions.
