Brief Details: 8B parameter LLaMA-based model optimized for creative writing & roleplay. Features 128K context, trained with diverse RP datasets & deduplication focus.
BRIEF DETAILS: SuperCorrect-7B is a 7.62B parameter LLM specializing in mathematical reasoning, achieving SOTA performance on MATH/GSM8K benchmarks through innovative two-stage fine-tuning.
Brief Details: SummLlama3-70B: A 70B parameter summarization model trained with DPO, optimized for human-preferred summaries across 7 domains.
Brief Details: SummLlama3-8B is an 8B parameter summarization model that outperforms Llama3-70B and GPT-4, specialized in human-preferred summaries across 7 domains.
Brief-details: CompassJudger-1-32B-Instruct is a comprehensive judge model based on Qwen2.5-32B, specializing in evaluation methods through scoring, comparison, and detailed assessment feedback.
BRIEF-DETAILS: PDF extraction toolkit model for document processing. Apache 2.0 licensed. Supports efficient PDF content extraction with HuggingFace integration.
Brief-details: FluxFit is an innovative virtual fitting model based on the Flux architecture, offering non-commercial virtual clothing try-on capabilities with ONNX and Safetensors support.
Brief-details: UI-focused multimodal LLM (8.4B params) built on Llama-3-8B, specialized for UI tasks with referring, grounding & reasoning capabilities
Brief Details: A 22B parameter creative language model built on Mistral architecture. Optimized for roleplay and creative writing with enhanced narrative capabilities.
Brief Details: Optimized ONNX version of Meta's Llama-3.2-3B-Instruct model for accelerated inference, supporting both CPU and GPU with int4 quantization, offering up to 39x speedup on A100 GPUs.
Brief Details: Deep Compression Autoencoder (DC-AE) model enabling high-resolution diffusion with 128x spatial compression while maintaining quality. 1.12B parameters, focuses on efficient image generation.
Brief-details: A 22B parameter merged language model combining Mistral-Small variants with Cydonia and Acolyte, optimized for creative text generation and storytelling capabilities in BF16 format.
Brief Details: Text-to-image prompt optimization model with 203M parameters, built on LLaMA architecture. Trained on multiple image captioning datasets for enhanced T2I generation.
Brief Details: A WebUI interface for music source separation with multi-language support, integrating MSST and UVR capabilities for audio processing and vocal removal.
Brief-details: NVIDIA's specialized math-focused 70B parameter LLaMA model, outperforming base LLaMA3.1 by 3.9% on MATH benchmarks with enhanced problem-solving capabilities.
BRIEF-DETAILS: 8B parameter LLaMA-based model with multiple GGUF quantizations optimized for different hardware setups. Features creative and instructional capabilities with extensive quantization options.
Brief-details: A 514M parameter image prompt enhancement model based on h2o-danube3-500m, specialized in generating detailed art descriptions with multiple prompt styles.
Brief Details: Korean-specialized LLaMA 3.1 variant (8.03B params) fine-tuned on 3.6GB of Korean cultural/educational data, supporting bilingual Ko-En tasks.
Brief Details: Advanced text-to-video synthesis model capable of generating high-quality videos at multiple resolutions (512-1024px). Part of the CogVideoX family with enhanced motion capabilities.
Brief Details: TIPO-500M: A 508M parameter LLaMA-based model for text-to-image prompt optimization. Trained on multiple datasets for enhanced image generation guidance.
Brief-details: Large 123B parameter language model fine-tuned on Claude-like datasets, optimized for high-quality prose generation with Mistral architecture and 16k context window
