Brief-details: 7.62B parameter multilingual LLM based on Qwen2, optimized through fine-tuning and DPO, achieving strong performance on MT-Bench and outperforming GPT-3.5 in various tasks.
BRIEF-DETAILS: A multi-modal transformer agent (193M params) trained across Atari, BabyAI, MetaWorld & MuJoCo environments, achieving strong performance in reinforcement learning tasks.
Brief Details: An 8B parameter LLaMA-3 variant optimized for multi-turn conversations, showing improved performance over base model with 8.33 MT-Bench score
Brief Details: A 12B parameter chat model by StabilityAI, featuring ChatML format support, function calling, and strong performance on benchmarks like MT-Bench (8.15 score).
BRIEF-DETAILS: Guanaco-65B is a powerful open-source chatbot built on LLaMA, fine-tuned using QLoRA on OASST1 dataset, competitive with commercial systems.
Brief-details: A powerful 20B parameter LLM with 200K context window, excelling in reasoning, math, and code. Strong performance comparable to GPT-3.5 in key benchmarks.
Brief Details: Aurora is a versatile Stable Diffusion model optimized for anime-style image generation, featuring advanced lighting control and multiple artistic styles including pixel art and semi-realistic outputs.
Brief Details: A 13B parameter medical LLM fine-tuned on diverse healthcare datasets including ChatDoctor, Wikidoc, and medical flashcards for Q&A tasks.
Brief Details: Fine-tuned 7B parameter Mistral model optimized for self-reflective RAG operations with 32k context window, featuring improved retrieval and generation capabilities.
Brief-details: COSMO-XL is a conversation agent built on T5, specialized in natural chitchat and social dialogue, trained on SODA and ProsocialDialog datasets for enhanced generalizability
Brief-details: A BART-based model fine-tuned on ChatGPT/GPT-4 prompts dataset, achieving 1.22 train loss with MIT license. Specialized in generating AI conversation prompts.
BRIEF DETAILS: DuckDB-NSQL-7B is a specialized SQL generation model based on Llama-2, fine-tuned on 200k DuckDB text-to-SQL pairs for accurate query generation.
Brief-details: AIM is Apple's advanced autoregressive image model collection offering scalable pre-training up to 7B parameters with strong ImageNet performance and multi-backend support (PyTorch/MLX/JAX).
Brief Details: MoMo-72B-LoRA-V1.4 is a 72.3B parameter LLM built on QWEN-72B, fine-tuned using LoRA technique with SlimOrca dataset. MIT licensed, optimized for text generation.
Brief-details: Multilingual embedding model supporting 100+ languages with strong performance on semantic search, clustering, and classification tasks. Features comprehensive MTEB benchmark results.
Brief-details: A specialized 13B parameter language model merging Xwin-LM with MLewd capabilities, designed for adult/NSFW content generation with enhanced creative storytelling abilities
Brief Details: Scratch-trained text-to-image diffusion model using public domain/CC0 images, built with OpenCLIP ViT-H/14 encoder, specialized for AI VTuber applications.
Brief Details: A 13B parameter Chinese NSFW instruction-tuned model based on Baichuan-13B-Base, featuring enhanced performance in text generation and CMMLU benchmarks.
Brief Details: A high-definition box-promptable object segmentation model designed for precise object cutouts at 1024x1024 resolution, with 94.6M parameters and MIT license.
Brief-details: A specialized Stable Diffusion diffusers model optimized for high-quality image generation, featuring FP32 precision fixes and NiPruned architecture
Brief-details: First open-source Chinese Stable Diffusion anime model trained on 1M curated anime image-text pairs. Features Chinese text prompts, anime-style generation, and built-in upscaling capabilities. 1B parameters.
