Brief Details: A fine-tuned BEiT model for pedestrian gender recognition achieving 91.07% accuracy, based on PETA dataset with 86.2M parameters.
BRIEF DETAILS: RoBERTa-based emotion recognition model trained on 58M tweets, capable of classifying emotions like joy, anger, sadness, and optimism with high accuracy.
BRIEF-DETAILS: Advanced 13B parameter language model merging MythoLogic-L2 and Huginn, optimized for roleplay and storytelling with unique tensor merging technique.
BRIEF DETAILS: Qwen2.5-14B is a powerful 14.8B parameter base LLM supporting 29+ languages with 128K context length, optimized for coding, mathematics, and long-form content generation.
Brief-details: High-quality Korean text-to-speech model from MyShell.ai's MeloTTS family, offering CPU real-time inference with MIT license and extensive language support.
Brief-details: Shuttle-3-Diffusion is an Apache-2.0 licensed text-to-image model capable of generating high-quality images in just 4 steps, featuring enhanced performance in typography and complex prompt understanding.
Brief-details: Chinese legal-focused LLaMA model (13B params) converted to GGUF format for efficient CPU/GPU inference via llama.cpp, optimized for legal domain tasks
Brief Details: A 6.97B parameter Mistral-based instruction model optimized for 4-bit inference, featuring enhanced memory efficiency and speed improvements.
Brief-details: A specialized tokenizer for LLaVA-Next architecture integrating Qwen and SigLIP components, designed for text generation and conversational AI tasks with transformer-based processing.
Brief-details: State-of-the-art monocular depth estimation model leveraging Stable Diffusion, offering zero-shot transfer capabilities for depth prediction from single images.
BRIEF DETAILS: ConvNeXt V2 tiny model with 28.6M params, specialized in image classification. Uses FCMAE framework and GRN layer. Apache 2.0 licensed.
Brief-details: Optimized 4-bit quantized Mistral-7B v0.3 model by Unsloth, offering 2.2x faster inference with 62% less memory usage. Apache 2.0 licensed.
Brief-details: A 14B parameter LLM built on Qwen2.5-14B-Instruct, featuring cross-architecture distillation from Llama-3.1-405B and Qwen2.5-72B models. Excels in instruction-following and reasoning.
Brief Details: Unsupervised contrastive learning BERT model for sentence embeddings, trained on Wikipedia data. Features strong alignment and uniformity properties for semantic similarity tasks.
Brief Details: SegFormer b2 encoder pre-trained on ImageNet-1k, designed for semantic segmentation with transformers. NVIDIA-developed with 22.8K+ downloads.
Brief Details: DistilBART model fine-tuned on MNLI task, achieving 88.1% matched accuracy through knowledge distillation, optimized for zero-shot classification.
Brief Details: A compact 2.43M parameter multimodal vision-language model based on Qwen1.5-0.5B, capable of image understanding and text generation with impressive benchmark scores.
BRIEF DETAILS: Text-to-image diffusion model merging VectorArtz and DucHaiten Lofi A styles, specialized in vector art and anime aesthetics with 23K+ downloads.
Brief Details: A photorealistic text-to-image model specialized in portrait photography, optimized for capturing detailed human subjects with cinematic quality.
Brief-details: LLaVA-NeXT-Video-7B-DPO is a multimodal video-text model with 7B parameters, capable of processing both video and image inputs for conversation and analysis.
Part 1 - Brief Details: Uncensored 8B parameter LLaMA 3.1-based conversational AI with 128K context window, optimized for instruction-following and coding tasks. Features ChatML format.
