Brief-details: Large-scale vision transformer model specialized in image classification, pre-trained on ImageNet-21k and fine-tuned on ImageNet-1k with improved architecture for better scaling and resolution handling.
BRIEF DETAILS: MT5-base model fine-tuned for Spanish text summarization, achieving 28.11 ROUGE-1 score. Built on wiki_lingua dataset with Apache 2.0 license.
Brief-details: A specialized transformer-based model for de-identifying radiology reports, achieving 97.9+ F1 scores across institutions. Built by Stanford AIMI for medical privacy.
Brief-details: A scikit-learn pipeline combining Hugging Face transformers with logistic regression for sentiment analysis, achieving 87% accuracy using BART embeddings.
Brief-details: Wav2vec2-based ASR model for Wolof language, trained on DVoice dataset. Achieves 16.05% WER on test set. Built with SpeechBrain framework.
Brief-details: DDIM is an efficient iterative implicit probabilistic model for fast, high-quality image generation, offering 10-50x speedup over DDPMs
Brief Details: A TensorFlow-based speaker recognition model using 1D CNN with residual connections, processes FFT-transformed audio at 16kHz for speaker classification.
Brief-details: DistilBERT model optimized for Apple Neural Engine, offering efficient text classification on Apple devices with SST-2 fine-tuning.
Brief-details: GLIDE is a powerful text-to-image diffusion model utilizing classifier-free guidance for photorealistic image generation and editing, licensed under Apache 2.0
BRIEF DETAILS: A TQC (Truncated Quantile Critics) reinforcement learning model trained on PandaReach-v1 environment, achieving -2.30 mean reward using stable-baselines3.
Brief-details: BERT-based keyword extraction model achieving 87% F1-score, fine-tuned on English text with strong precision (85%) and recall (88%) metrics
Brief-details: LayoutLMv3 model fine-tuned for invoice data extraction, achieving perfect accuracy in identifying key invoice fields like biller details, dates, and amounts.
Brief-details: TQC (Truncated Quantile Critics) reinforcement learning model trained on FetchPickAndPlace-v1 environment, achieving -8.50 mean reward using stable-baselines3.
Brief-details: A specialized text-to-text generation model trained on multiple data-to-text tasks, optimized for converting structured data into natural language descriptions.
Brief-details: PPO reinforcement learning model trained on HalfCheetah-v3 environment, achieving mean reward of 5836.27. Built with stable-baselines3, optimized for locomotion tasks.
Brief-details: BERT model fine-tuned for banking-related text classification, achieving 92.76% accuracy on Banking77 dataset. Optimized for customer service inquiries.
Brief-details: RoBERTa-based masked language model for Catalan, trained on 35GB of text data. Strong performance on NLP tasks like NER, QA, and classification.
Brief Details: Chinese speech model pretrained on 10k hours WenetSpeech, designed for feature extraction and audio processing. MIT licensed, 4.4k+ downloads.
Brief Details: T5-large model fine-tuned for grocery domain question generation, achieving 91.39 BERTScore. Built on SubjQA dataset with comprehensive evaluation metrics.
Brief Details: MobileViT + DeepLabV3 small model for semantic segmentation, combining transformer-based vision processing with 6.4M params, achieving 79.1% mIOU on PASCAL VOC.
Brief-details: BLOOM-560M is a 559M parameter multilingual language model supporting 48 languages, trained by BigScience. Optimized for text generation with FP16 precision.
