Brief Details: JoyVASA is a diffusion-based AI model for generating facial animations from audio, supporting multilingual inputs and capable of animating both human and animal faces.
Brief Details: Llama-3.1-Tulu-3-8B is an 8B parameter instruction-following model optimized for diverse tasks including MATH and GSM8K, with strong performance in reasoning and safety.
BRIEF DETAILS: Marco-o1 is a 7.62B parameter reasoning-focused LLM built on Qwen2, featuring Chain-of-Thought fine-tuning and Monte Carlo Tree Search for enhanced problem-solving capabilities.
BRIEF-DETAILS: A specialized LoRA model trained on Namsan Tower images, built on FLUX.1-dev base model for generating Seoul landmark imagery with 47 likes and 275 downloads.
Brief-details: A specialized LoRA model for generating Korean Hanbok traditional clothing images, built on FLUX.1-dev base model with non-commercial license support and diffusers integration.
Brief-details: EchoMimicV2 is an advanced AI model for creating striking audio-driven human animations, supporting both English and Chinese audio inputs with semi-body motion generation.
Brief-details: MobileCLIP is a fast, efficient image-text model from Apple that achieves SOTA performance while being significantly smaller and faster than competitors
Brief-details: Fine-tuned 8B parameter Llama-3.1 model specialized for scientific literature synthesis, developed by UW and AI2, trained on os-data dataset
Brief-details: A specialized FLUX-based LoRA model focused on Rolls-Royce car image generation, supporting both English and Korean prompts with high-quality automotive visualization capabilities.
Brief-details: Image classification model for real estate photos achieving 89.58% accuracy. Classifies rooms and building facades using Vision Transformer architecture.
BRIEF-DETAILS: CodeBERT model fine-tuned on C code for masked language modeling, designed for code evaluation and analysis with 1M training steps
Brief Details: A lightweight Chinese GPT-2 model developed by CKIPLAB for traditional Chinese text generation, featuring transformer architecture and GPT-2 capabilities with specialized tokenization requirements.
Brief-details: Speech emotion recognition model fine-tuned on wav2vec2, achieving 97.46% accuracy across 7 emotions using SAVEE, RAVDESS, and TESS datasets.
Brief Details: A compact BERT model fine-tuned for intent classification on the MASSIVE dataset, achieving 85.34% accuracy using knowledge distillation
Brief Details: A 320M parameter Chinese DeBERTa-v2 model trained on WuDao Corpora, specialized in NLU tasks with whole word masking. Strong performance on AFQMC, TNEWS, IFLYTEK benchmarks.
BRIEF DETAILS: Indonesian BERT-based sentence embedding model that maps text to 768-dimensional vectors, optimized for semantic similarity tasks in Bahasa Indonesia.
Brief-details: German BERT model specialized in legal named entity recognition, supporting 18 entity classes for German legal texts. 109M params, PyTorch-based.
Brief-details: A 2B parameter Verilog code generation model fine-tuned from CodeGen-multi-2B, specialized for hardware description language generation with 8-day training on Tesla A100 GPUs
Brief-details: Text generation model trained on @perpetualg00se's tweets using GPT-2, featuring 2,024 curated tweets and specialized for Twitter-style content generation.
BRIEF DETAILS: RoBERTa-based text classification model fine-tuned on SST2 dataset, achieving 93.23% accuracy for sentiment analysis. MIT licensed with PyTorch implementation.
Brief-details: A Stable Diffusion model fine-tuned on Pokémon images with BLIP captions, capable of generating custom Pokémon from text prompts. Trained by Lambda Labs.
