Brief-details: Powerful 72B parameter chat model quantized to 4.65 bits-per-weight, rivals GPT-4o performance, supports 32K context, excellent at chat/math/coding tasks
Brief-details: CraftsMan is an advanced image-to-3D mesh generation model focusing on high-fidelity output with native 3D generation and interactive geometry refinement capabilities.
Brief-details: Windows-compatible wheel builds for flash-attention library, supporting efficient CUDA-based attention computation with BSD-3-Clause license
Brief Details: CamemBERTav2-base is a French language model with 111M parameters, trained on 275B tokens using DebertaV2 architecture with MIT license.
Brief Details: OS-Atlas-Pro-7B is an advanced 8.29B parameter GUI action model built on Qwen2-VL-7B-Instruct, specialized in generating thoughtful reasoning and actions for GUI tasks.
Brief Details: Vietnamese-English visual language model with 938M parameters, optimized for document understanding and RAG capabilities using Colpali architecture.
Brief Details: A powerful 13.9B parameter text-to-image model using MMDiT architecture with multiple CLIP encoders and QK normalization, offering enhanced image quality and text understanding.
Brief Details: A powerful 32B parameter code-specialized LLM with 4-bit AWQ quantization, offering enhanced code generation, reasoning, and fixing capabilities with 128K token context.
Brief Details: Qwen2.5-Coder-3B is a specialized code-focused LLM with 3.09B parameters, offering 32K context length and enhanced code generation capabilities.
Brief-details: A 72B parameter instruction-tuned language model quantized to 2-bit precision using AutoRound, offering efficient deployment while maintaining performance.
Brief-details: Experimental HDR-focused LoRA model for FLUX.1-dev, optimized for 1024x1024 images with AdamW optimizer and constant LR scheduler. Features specialized HDR image generation capabilities.
Brief Details: Multimodal Speech LLM combining Mistral-Nemo and Whisper for speech/text processing. 52.4M params, supports 15 languages, MIT license.
Brief-details: An 8B parameter multimodal LLM with enhanced reasoning capabilities through Mixed Preference Optimization (MPO), achieving strong performance on visual-language tasks.
Brief-details: A 72.7B parameter fine-tuned Qwen2.5 model optimized for roleplay/creative writing, trained on diverse datasets with improved instruction following and context understanding
Brief-details: High-performance text/image-to-video generation model with 12B parameters, supporting multiple resolutions and languages, featuring advanced control capabilities and efficiency optimizations.
Brief Details: A 9B parameter multilingual instruction-tuned LLM optimized for Indonesian, Javanese, and Sundanese languages, built on Gemma2 architecture
Brief Details: A 14.8B parameter story-writing model built on SuperNova-Medius, optimized for prose generation with enhanced context handling and instruction following capabilities.
BRIEF DETAILS: A 10.2B parameter multilingual LLM focused on Indonesian, Javanese, and Sundanese languages, built on Gemma2-9B with excellent regional language performance.
Brief Details: A specialized geospatial AI model for predicting tree and vegetation canopy heights using satellite imagery, built on Swin-B transformer architecture.
Brief-details: LLaMa-based TTS model with 350M parameters, offering pure language modeling approach for speech synthesis and voice cloning capabilities, optimized for English text-to-speech conversion.
Brief Details: OmniGen-V1 is a 3.88B parameter unified image generation model capable of multi-modal generation without plugins, supporting text-to-image, subject-driven, and image editing tasks.
