Brief-details: Monolingual Slovak RoBERTa-base model trained on 4.5GB of cleaned news data, optimized for Slovak language processing tasks
Brief-details: A robust speech recognition model fine-tuned on Switchboard telephone data, pre-trained on multiple speech corpora, optimized for 16kHz audio transcription.
Brief-details: DeiT-base-patch16-384 is a data-efficient Vision Transformer with 87M parameters, achieving 82.9% top-1 accuracy on ImageNet, optimized for 384x384 resolution images
Brief-details: A 90M parameter open-domain chatbot developed by Facebook, focused on engaging multi-turn dialogue with enhanced conversational skills and human-like interactions.
BRIEF DETAILS: Vietnamese transformer-based NLP model for UD v2.5, offering high-accuracy tokenization (98.42%) and POS tagging (90.19%) with comprehensive linguistic analysis capabilities.
Brief-details: StyleGAN2-based anime character generator creating 256x256 full-body female anime characters with white backgrounds. Trained for 150 epochs using stylegan2-pytorch library.
Brief Details: Fine-tuned distilGPT2 model optimized for conversational AI, trained on Wizard of Wikipedia dataset with persona framework for chatbot applications. Loss: 2.2461.
BRIEF-DETAILS: Classical Chinese sentence segmentation model by ethanyt. Specialized tool for processing ancient Chinese texts with automated segmentation capabilities.
BRIEF-DETAILS: ESPnet2 ASR Conformer model trained on SEAME dataset, utilizing BPE tokenization with 5626 units for end-to-end speech recognition.
Brief-details: ESM-1v is a 650M parameter protein language model by Facebook, trained on UniRef90 for protein sequence analysis and representation learning
Brief-details: Vision Transformer (ViT) model trained on LAION-400M dataset, featuring large-scale architecture with 14x14 patches and CLIP training compatibility
Brief Details: A minimal GPT-2 language model implementation with randomly initialized weights, designed for testing and educational purposes
Brief Details: A re-ranking cross-encoder model based on mMiniLMv2, featuring L12-H384 architecture, designed for multilingual document retrieval tasks
BRIEF-DETAILS: DeBERTa-large model fine-tuned for formal vs informal text classification, achieving 87.8% accuracy and strong F1 scores for both styles.
Brief-details: Quantized versions of Gemma 2B model optimized for various hardware configurations, offering multiple compression levels from 1.37GB to 10.46GB with imatrix calibration
Brief Details: BakLlava-v1-hf is a Mistral-7B-based multimodal model combining LLaVA 1.5 architecture for vision-language tasks, featuring enhanced performance over Llama 2 13B.
Brief-details: Japanese sentiment analysis model based on LUKE, fine-tuned to classify 8 emotions (joy, sadness, anticipation, surprise, anger, fear, disgust, trust) using the WRIME dataset.
Brief-details: Wav2Vec2-based emotion recognition model trained on IEMOCAP dataset. Achieves 62.58% accuracy for 4-class emotion classification from 16kHz speech.
Brief-details: A lightweight CodeGen-based causal language model designed for code generation tasks, maintained by Hugging Face's private model repository
Brief-details: Parakeet CTC 1.1B is a large-scale ASR model (1.1B parameters) for English speech transcription, using FastConformer architecture with CTC loss, trained on 64K hours of speech data.
BRIEF-DETAILS: 4-bit quantized version of Qwen2.5-32B-Instruct optimized for MLX framework, offering efficient deployment with reduced memory footprint
