BRIEF-DETAILS: State-of-the-art small embedding model (281M params) achieving 65.0 MTEB score through innovative contextual document embedding approach
Brief-details: StarChat-β is a 15.5B parameter coding assistant model fine-tuned from StarCoderPlus, optimized for programming tasks and chat interactions across 80+ programming languages.
Brief Details: NSFW-gen-v2 is an advanced text-to-image generator by UnfilteredAI, featuring 3D rendering capabilities and FP16 optimization, designed for explicit content generation.
Brief Details: A natively fine-tuned version of Stanford's Alpaca model trained on 4xA100s, achieving 41.96 avg score on OpenLLM benchmark. Built on LLaMA architecture.
Brief-details: Mixtral-7B-8Expert is a multilingual MoE (Mixture of Experts) model supporting 5 languages, built on Mistral architecture with strong performance across multiple benchmarks
Brief Details: Llama-3 8B parameter model optimized for tool use and function calling. Features 89.06% BFCL accuracy. Built on Meta's base model.
Brief Details: RWKV-5 World is a multilingual language model trained on 1.12T tokens across 12 languages, optimized for text generation and coding tasks with PyTorch implementation.
BRIEF DETAILS: Advanced 73.3B parameter multimodal AI model from Allen AI that excels at image-text tasks, achieving top academic benchmark scores and near GPT-4 performance.
BRIEF-DETAILS: AI model optimized for realistic Asian-style image generation, based on merged Basilmix. Uses vae-ft-mse-840000 and includes Ulzzang-6500 embeddings.
Brief-details: ChatGLM-6B RLHF-enhanced model trained on 700k+ high-quality data samples, optimized for Chinese/English dialogue with unlimited context length and improved summarization capabilities.
Brief-details: A distilled BART model trained on CNN/DailyMail dataset for text summarization, offering 1.24x speedup with comparable ROUGE scores to BART-large-cnn
Brief-details: LivePortrait is an efficient portrait animation model that enables high-quality face reenactment with stitching and retargeting control mechanisms, supporting both image and video inputs.
Brief-details: Llama2-Chinese-13b-Chat is a specialized Chinese-optimized version of Meta's Llama-2 model, fine-tuned using LoRA for enhanced Chinese language capabilities while maintaining English proficiency
Brief-details: A 34B parameter instruction-tuned Code Llama model optimized for code generation and understanding, featuring chat capabilities and Meta's advanced architecture.
BRIEF-DETAILS: Open-Assistant's 12B parameter English language model, fine-tuned on 22k human demonstrations. Built for assistant-style conversations with strong text generation capabilities.
BRIEF DETAILS: Massive 1.6T parameter Mixture-of-Experts model trained on C4 dataset for text generation, featuring 2048 experts and 4x speedup over T5-XXL.
Brief Details: Fine-tuned Stable Diffusion model trained on "Loving Vincent" film screenshots, creating Van Gogh-style images using 'lvngvncnt' token.
Brief Details: 3B parameter code completion model with 4k context window, trained on diverse programming languages. Achieves 17.68% pass@1 on HumanEval.
Brief Details: DeepSeek-V2 is a 236B parameter MoE model with 21B active parameters per token, offering exceptional efficiency and 128k context length
BRIEF-DETAILS: A 350M parameter text-to-speech model built on LLaMa architecture, offering voice cloning capabilities and pure language modeling approach without external adapters.
Brief-details: An uncensored 13B parameter LLaMA-based model trained on filtered Wizard-Vicuna data, designed for unrestricted text generation with strong performance on various benchmarks.
