Brief-details: State-of-the-art 7B math model achieving 59.1% on AIME24, trained with only 3K data. Built on DeepSeek-R1-Distill-Qwen-7B with exceptional mathematical reasoning capabilities.
Brief-details: MIDI-3D is a revolutionary 3D generative AI model that transforms single images into complete 3D scenes using multi-instance diffusion techniques, developed by VAST-AI.
Brief-details: A 3B parameter LLM with unique dual-mode capability - combines traditional chat responses with deep reasoning chains, featuring advanced function calling and JSON output modes
BRIEF-DETAILS: A powerful reranking model supporting 100+ languages with SOTA performance, optimized for search and retrieval tasks. Features multilingual & code support.
BRIEF-DETAILS: A 72B parameter distilled LLM optimized for mathematical and general reasoning, showing strong performance across benchmarks with 5-11% improvements over comparable models.
Brief-details: YOLO World Mirror provides Ultralytics YOLO model weights for object detection, offering efficient implementation of YOLO architecture with comprehensive documentation support.
Brief-details: ControlNet modules in .safetensors format for stable-diffusion-webui, offering pruned versions of lllyasviel's ControlNet and TencentARC's T2I-Adapters
BRIEF DETAILS: LLaMA.cpp-optimized quantized versions of Google's Gemma 3B model with vision capabilities, offering multiple compression levels from 23GB to 4GB
Brief Details: EuroBERT-2.1B is a powerful multilingual encoder supporting 15 languages with 2.1B parameters, optimized for retrieval, classification & code tasks.
Brief Details: Steiner-32b is an experimental reasoning model focused on autonomous exploration of multiple reasoning paths, trained via RL on synthetic data
Brief Details: DeepHermes-3 is a 3B parameter hybrid reasoning LLM that uniquely combines intuitive responses and chain-of-thought reasoning, built on Llama 3.
Brief Details: Light-R1-14B-DS is a 14B parameter SOTA math model achieving impressive AIME scores (74.0/60.2), featuring successful RL implementation on long-COT finetuned models.
Brief Details: A multimodal LLM combining SigLIP vision encoder & Phi-3.5-mini LLM, featuring novel modality-mutual attention for enhanced vision-language alignment
Brief-details: LLMVoX is a 30M-parameter streaming text-to-speech model designed for LLM integration, offering low-latency speech synthesis with multi-queue streaming capabilities.
BRIEF-DETAILS: Uncensored 27B parameter Gemma-IT model using layerwise abliteration technique, maintaining >90% acceptance rate while preserving core capabilities.
Brief Details: A specialized LoRA model for Wan2.1 14B I2V that enables 360-degree rotation effects in video generation, trained on rotation-specific data for 20 epochs.
Brief-details: A specialized LoRA model trained on Steamboat Willie clips for generating golden era animation-style videos, compatible with Wan2.1-T2V models and featuring distinct vintage cartoon aesthetics.
Brief Details: A 32B parameter LLM merged using TIES method, combining Qwen2.5-32B variants with strong reasoning and roleplay capabilities. Notable for minimal hallucination and consistent character portrayal.
BRIEF-DETAILS: Gemma 3B (12B parameter variant) pretrained model from Google, requiring acceptance of usage license. Advanced language model with state-of-the-art capabilities.
Brief-details: OLMo-2-0325-32B is a 32B parameter open-source language model from Allen AI, trained on 6T tokens with strong performance across various benchmarks.
BRIEF-DETAILS: Multilingual BERT model with 210M parameters supporting 15 languages. Optimized for retrieval, classification, and regression tasks with 8K token sequences.
