Brief Details: Meta's latest 70B parameter instruction-tuned LLM, built for advanced natural language processing and generation tasks, with enhanced instruction-following capabilities.
Brief-details: Zero-shot topic classifier fine-tuned on Yahoo Answers, built on BART-MNLI architecture for flexible label predictions without prior training
Brief-details: RT-DETR real-time object detector combining DETR & YOLO capabilities. Achieves 53.1% AP on COCO at 108 FPS. Pre-trained on COCO and Objects365 datasets.
Brief Details: A specialized speech embedding model combining pyannote framework with WeSpeaker architecture using ResNet34, trained on VoxCeleb dataset for speaker recognition tasks.
BRIEF DETAILS: A powerful text reranker model (149M params) from Alibaba-NLP, built on modernBERT architecture with 8192 token support and strong performance on BEIR/LoCo benchmarks.
BRIEF-DETAILS: 4-bit quantized version of Microsoft's Phi-4 model optimized for MLX framework, offering efficient deployment with reduced memory footprint.
BRIEF-DETAILS: Qwen2.5-1.5B-Instruct-4bit is a 4-bit quantized version of Qwen 2.5 (1.5B parameters) optimized for MLX framework, offering efficient instruction-following capabilities.
Brief Details: Zero-shot text classification model optimized in ONNX format, based on MNLI architecture from Facebook. Efficient for deployment and inference.
BRIEF-DETAILS: A specialized CANINE-based model designed for text processing with WTPSplit, featuring a single-layer architecture optimized for efficient text analysis.
Brief-details: AraBERT - Arabic BERT model with 136M parameters. Pre-trained on 77GB Arabic text. Supports both segmented and non-segmented text processing. Optimized for Arabic NLP tasks.
BRIEF-DETAILS: Qwen2.5-Coder-14B-Instruct-4bit is an MLX-optimized coding model, converted from Qwen's original 14B parameter model with 4-bit quantization for efficient deployment.
Brief Details: A compact implementation of the MinicPMv2.6 architecture by katuni4ka, designed for efficient processing with randomized parameters
Brief-details: Mistral-7B-v0.2 is a powerful 7B parameter language model from the Mistral AI community, featuring improved capabilities over its predecessor and designed for various NLP tasks.
BRIEF-DETAILS: A lightweight 0.18B parameter LLM optimized for CPU inference, capable of 160 tokens/sec on single core. Features GrokAdamW optimizer and evolutionary merging.
BRIEF-DETAILS: Meta's Sapiens: Advanced vision model family for human analysis - pose estimation, segmentation, depth & surface normal prediction. Scales 0.3B-2B params.
Brief Details: Meditron-70B is a specialized medical LLM with 70B parameters, built on Llama-2, trained on extensive medical literature and guidelines for enhanced clinical reasoning.
Brief Details: CosXL is a novel AI model from StabilityAI emphasizing controlled text generation with advanced architecture and versatile applications.
BRIEF-DETAILS: OLMoE-1B-7B-0125-Instruct: Advanced mixture-of-experts LLM with strong performance on math, coding & reasoning tasks. Apache 2.0 licensed, primarily English focused.
Brief Details: YuE-s1-7B-anneal-en-icl is an open-source music generation model that transforms lyrics into complete songs with vocals and accompaniment, licensed under Apache 2.0.
BRIEF DETAILS: Uncensored 70B parameter LLaMA model variant created through abliteration technique to remove refusal responses, built on DeepSeek's base model.
Brief Details: Japanese-optimized 7B parameter LLM based on Qwen2.5, fine-tuned with DPO on synthetic Japanese conversation data. Outperforms base model by 16.2% on multilingual benchmarks.
