Brief-details: A 1.3B parameter SQL generation model that outperforms larger models on text-to-SQL tasks, featuring impressive accuracy on easy/medium/hard queries and Apache 2.0 license.
Brief-details: GPL-licensed 13B parameter LLaMA-based chatbot by Nomic AI, achieving 65.3% avg score on reasoning benchmarks. Excels at assistant-style interactions.
BRIEF DETAILS: Hybrid 1.5B parameter language model combining Mamba and Attention heads, outperforming sub-2B models. Features meta tokens and shared KV cache for efficiency.
Brief Details: Experimental AI model merging 4 teacher models using attention agreement scoring. Optimized for 2048x2048 resolution with specialized workflow requirements.
Brief-details: Atom-7B-Chat is a bilingual Chinese-English LLM with 7B parameters, featuring enhanced context length (32k), optimized Chinese vocabulary, and commercial usage rights under Apache 2.0.
Brief-details: InternLM2.5-20B-Chat is a powerful 20B parameter LLM with exceptional math reasoning capabilities, surpassing Llama3 and Gemma2-27B in benchmarks.
BRIEF-DETAILS: Double Exposure embedding for SD 2.x - Creates artistic double exposure effects with portraits and landscapes. Trained on 768px images for v2.1.
Brief-details: DeepSeek-Coder-V2-Lite-Instruct-GGUF is a 15.7B parameter coding-specialized LLM with multiple GGUF quantizations for efficient deployment
Brief-details: MC-LLaVA-3b is a 3.22B parameter vision-language model using multi-crop image processing, fine-tuned from Phi-2 with SigLIP vision tower. Achieves 76.72% on VQAv2.
Brief-details: A 33B parameter cybersecurity-focused LLM built on DeepSeek architecture, specializing in offensive and defensive security analysis with enhanced prompting capabilities.
Brief Details: Bunny-Llama-3-8B-V is a lightweight multimodal model with 8.48B parameters, combining SigLIP vision encoder and Llama-3-8B for efficient image-text tasks.
Brief-details: State-of-the-art 70B parameter LLM fine-tuned on 300k+ instructions, featuring improved response length and reduced hallucination rate
Brief-details: SDXL DPO Turbo is a merged text-to-image model combining SDXL Turbo & SDXL DPO capabilities, offering fast inference with just 8 steps.
Brief-details: GPTQ-quantized version of Meta's Llama-2-70B model optimized for efficient deployment. Features 4-bit and 3-bit quantization options with various group sizes for VRAM optimization.
Brief-details: Dolphin 2.9 Llama3 8B - Uncensored conversational AI model with 8B parameters, supporting coding, instruction-following, and function calling. Built on Meta's Llama 3.
Brief-details: Karlo v1-alpha is a sophisticated text-to-image generation model using unCLIP architecture, featuring improved super-resolution capabilities and efficient processing in minimal denoising steps.
BRIEF DETAILS: 8B parameter GGUF model merging Hermes 2 Pro and Llama-3, optimized for instruction-following and function calling with strong benchmark scores
Brief-details: A 20B parameter language model fine-tuned for adult content generation, built on GPT-NeoX architecture with comprehensive training on carefully curated NSFW datasets.
Brief Details: Chinese Alpaca-2 13B is a bilingual instruction-tuned LLM based on LLaMA-2, optimized for Chinese language tasks with extended vocabulary and 4K context window.
Brief Details: Phi-2-super: A 2.78B parameter fine-tuned language model based on Microsoft's Phi-2, optimized with SFT + cDPO for enhanced instruction following and chat capabilities.
Brief-details: Text-to-image diffusion model specialized in generating high-quality pony artwork, fine-tuned on 80k curated images with derpibooru tags. Built on waifu-diffusion.
