Brief Details: A 32B parameter math-focused model achieving near-SOTA performance on AIME benchmarks, trained with only 3K data points from DeepSeek base
Brief-details: Optimized GGUF quantizations of Google's Gemma-3B-it model with vision capabilities, offering various compression levels and specialized formats for different hardware.
Brief-details: AI model that transforms images into videos showing objects being cut open to reveal cake interiors. Built on Wan2.1 14B I2V 480p base model.
Brief Details: Enhanced Long-CLIP model with 248 token input capacity, featuring register tokens and gated MLPs. Significantly reduces modality gap and improves retrieval performance.
BRIEF DETAILS: Enhanced CLIP model with register tokens and gated MLPs, offering improved modality gap reduction and better performance across vision tasks. +20M params over standard CLIP.
BRIEF-DETAILS: Multilingual AI model supporting 23 languages with 256k context window, optimized for conversation, RAG, coding & tool use. GGUF quantized version.
Brief Details: A LoRA model trained on Replicate for image generation, using TOK as trigger word. Built for use with diffusers library and Flux trainer.
Brief-details: VRAM-24 is a specialized AI model by unslothai, focused on optimizing VRAM usage for deep learning applications, available on HuggingFace.
Brief-details: A computer vision model for binary gender classification (male/female) created using HuggingPics. Autogenerated image classifier suitable for gender detection tasks.
Brief-details: C4AI Command-R v01 - A Cohere research model focused on command processing and natural language understanding. Part of CohereForAI's research initiatives.
Brief Details: Muchi is a finetuned speech-text foundation model enabling real-time dialogue with 200ms latency, built on Moshi's architecture with enhanced conversational coherence and speech quality.
Brief Details: Advanced 24B parameter LLM with unique dual-mode reasoning capabilities, quantized for efficient inference and built on Mistral architecture
Brief-details: YOLOE is a highly efficient real-time object detection and segmentation model that supports text prompts, visual inputs, and prompt-free paradigms, achieving state-of-the-art performance.
BRIEF DETAILS: Gemma-3 1B instruction-tuned model in GGUF format, optimized by Unsloth. Part of Google's Gemma family, supports 140+ languages with 32K context window.
Brief-details: RWKV7-G1 "GooseOne" is a pure RNN reasoning model developed by BlinkDL, featuring advanced reasoning capabilities and efficient inference through the RWKV architecture.
Brief-details: MetaStone-L1-7B is a 7B parameter lite reasoning model achieving SOTA results in mathematics and coding tasks, comparable to Claude-3.5 and GPT4, built on DeepSeek-R1-Distill-Qwen-7B.
Brief-details: Unspecified AI model from unslothai available on HuggingFace. Limited information available about architecture and capabilities. Further details pending documentation.
BRIEF-DETAILS: A minimal Qwen2 causal language model designed specifically for TRL library testing purposes, focused on efficient unit testing.
Brief Details: DBRX-Instruct is Databricks' instruction-tuned language model designed for enterprise use, with privacy-aware data handling and processing capabilities.
Brief Details: Falcon-180B is TII's advanced 180-billion parameter language model, released with specific licensing requirements and use policies.
Brief-details: A comprehensive collection of GGUF quantized versions of the Deductive-Reasoning-Qwen-32B model, offering various compression levels from 65GB to 10GB with different quality-size tradeoffs
