Brief-details: A 6.7B parameter code LLM achieving state-of-the-art performance on code tasks, reproducing Microsoft's WarriorCoder with open-source models
Brief Details: SkyReels I2V Smooth LoRA - A specialized image-to-video LoRA model focused on creating smooth video transitions from images on Hugging Face.
Brief-details: A specialized AI model developed by Conard for fortune telling and prediction tasks, hosted on Hugging Face. Aimed at generating fortune-based predictions.
Brief-details: FastText-based binary classifier for high-quality data filtering, used to build PreSelect-100B dataset with 10% selection threshold. Developed by HKUST-NLP team.
Brief-details: SigLIP 2 is Google's advanced vision-language model trained on WebLI dataset, featuring improved semantic understanding and localization capabilities using TPU-v5e chips.
Brief Details: Fine-tuned 24B parameter Mistral model optimized for mathematical reasoning, achieving 95% accuracy on MATH-500 and strong AIME performance
Brief-details: A 72B parameter vision-language model optimized with AWQ quantization, featuring enhanced visual understanding, video processing up to 1+ hour, and structured output generation capabilities.
Brief-details: Phidias-Diffusion is a groundbreaking generative AI model for creating 3D content from text, images, and 3D conditions using reference-augmented diffusion techniques.
Brief-details: Ovis2-34B is a powerful multimodal LLM combining visual and textual capabilities, featuring enhanced reasoning, video processing, and multilingual OCR support with 34B parameters.
BRIEF DETAILS: A 24B parameter Mistral-based language model developed by ArliAI, optimized for role-playing and conversational tasks with enhanced contextual understanding.
BRIEF-DETAILS: A 32B parameter successor to s1 model with enhanced reasoning capabilities, achieving 95.4% on MATH500 and showing strong performance on AIME tests
BRIEF-DETAILS: Advanced AI autocorrection & enhancement engine for GGUF models - improves text generation quality, fixes common issues, and optimizes low-quantized models.
BRIEF-DETAILS: Japanese-optimized 14B parameter LLM, based on DeepSeek-R1-Distill-Qwen. Features MIT license and specialized Japanese language capabilities.
Brief Details: Multi-modal 7B parameter LLM capable of processing text, images, video & audio. Built on Qwen2.5 with 32K context window. Specializes in flexible visual processing.
Brief-details: OLMoE-1B-7B-0125 is an efficient Mixture-of-Experts LLM with 1B active/7B total parameters, achieving SOTA performance comparable to Llama2-13B.
Brief-details: A development-only debug model for Chroma implementations, created by lodestones. Currently in research phase with planned Apache 2.0 license release.
BRIEF DETAILS: 4-bit quantized version of DeepSeek-R1 model optimized for MLX framework, offering efficient deployment with reduced memory footprint
Brief-details: DeepSeek-R1-Distill-Llama-70B-GGUF is a distilled version of the DeepSeek-R1 model, based on Llama 3.3-70B-Instruct, optimized for reasoning tasks.
Brief-details: DeepSeek-R1-Distill-Llama-8B-GGUF is an 8B parameter distilled version of DeepSeek-R1, optimized for reasoning tasks using Llama 3.1 architecture with GGUF format support.
Brief Details: ToriiGate-v0.4-7B is a specialized vision-language model for anime/artwork captioning, built on Qwen2-VL with 900k+ training samples and advanced character recognition capabilities.
Brief-details: Mathematical premise retrieval model for Lean theorem proving, featuring pre-trained models for retrieval and reranking tasks. Developed by ruc-ai4math team.
