Brief Details: Llama3-8B model fine-tuned on BitNet 1.58b architecture, optimized for extreme quantization with 2.8B parameters. Trained on FineWeb-edu dataset reaching 100B tokens.
BRIEF DETAILS: 120B parameter scientific language model by Facebook, trained on 106B tokens of scientific text for research tasks like citation prediction and mathematical reasoning.
Brief-details: Improved autoencoder model fine-tuned on LAION-Aesthetics, optimized for Stable Diffusion with better face reconstruction and EMA weights implementation.
BRIEF DETAILS: LCM-SDXL is a high-performance text-to-image model that enables fast image generation in just 2-8 inference steps, based on SDXL with Latent Consistency optimization.
Brief-details: YOLOv8 is Ultralytics' latest SOTA model for object detection, segmentation, and classification, offering superior speed and accuracy with multiple pre-trained variants
Brief Details: A Chinese ancient-style text-to-image model focused on 2.5D aesthetic character generation, with improved face generation and easier prompting compared to v1.
BRIEF DETAILS: Specialized Stable Diffusion model fine-tuned for generating voxel art images, featuring creative ML open rail license and diffusers pipeline support.
Brief-details: Advanced conversational AI model based on Mixtral-8x7b architecture, optimized for coding and general tasks with 16k context window and uncensored capabilities
Brief Details: RWKV-4-Pile-7B is a 7B parameter language model trained on The Pile dataset, featuring 32 layers, 4096 embedding size, and Apache 2.0 license. Supports 1024-4096 context length.
Brief Details: NeuralBeagle14-7B is a 7.24B parameter DPO-tuned language model ranking first in its size category, featuring 8k context window and strong reasoning capabilities.
Brief-details: WizardLM 7B GGML is a CPU/GPU-optimized model offering multiple quantization options (4-bit to 8-bit), ideal for efficient local deployment with llama.cpp compatibility.
Brief-details: LoRA adaptation weights for Stable Diffusion 2.1, focused on high-quality image generation with Midjourney-like aesthetics, 157 likes and 118 downloads.
Brief-details: A 3.43B parameter instruction-tuned LLaMA model trained on WizardLM, Alpaca & Dolly datasets using Orca paper approaches, achieving 39.03% avg benchmark score.
Brief Details: IP-Adapter integration for FLUX.1-dev model using SiglipVisionModel. Enables image-guided text-to-image generation with 128 image tokens and MLPProjModel architecture.
Brief-details: Long-context instruction-tuned LLaMA2 variant with 32K context window, optimized for chat, summarization & QA tasks. Built on Together API.
Brief-details: A 70B parameter LLM optimized for creative writing and roleplay, merging Midnight-Miqu-70B-v1.0 with Tess-70B-v1.6 using DARE linear merge method.
Brief Details: FLUX.1-dev-Controlnet-Canny is a specialized ControlNet model for edge detection and image synthesis, trained on 1024x1024 resolution with 30k steps
Brief-details: Powerful 33B parameter coding model optimized for software development. Features 87% code/13% language training data, 16K context window, and multiple quantization options.
Brief Details: A specialized LoRA model for FLUX.1-dev that creates unique cartoon-realistic fusion images, featuring illustrated characters in realistic environments.
BRIEF DETAILS: 72B parameter LLM fine-tuned on Qwen2-72B-Instruct, optimized for Claude 3-like prose quality. Strong 76.06% IFEval accuracy, supports English/Chinese.
Brief-details: First open-source 1024x576 text-to-video model trained on 2,197 clips with 68,388 tagged frames using BLIP2. Built for high-resolution video generation.
