Brief Details: Large Vision Transformer (ViT) with 304M params, CLIP-pretrained on WIT-400M, fine-tuned on ImageNet-12k & ImageNet-1k. Excellent for image classification.
Brief-details: Faroese speech recognition model achieving 7.6% WER on test data, fine-tuned from wav2vec2-large-xlsr-53 using 100 hours of Faroese audio data.
Brief-details: Multilingual BERT-based sarcasm detector for news headlines in English, Dutch, and Italian. Achieves 87.23% F1 score with strong cross-language performance.
Brief-details: Large-scale Vietnamese speech model (317M params) trained on 13k hours of YouTube audio for self-supervised learning using wav2vec2 architecture.
Brief Details: A Chinese offensive language detection model based on RoBERTa, achieving 82.75% accuracy. Fine-tuned on COLDataset with 102M parameters for text classification.
BRIEF-DETAILS: PyTorch-based GAN library offering comprehensive implementations of conditional/unconditional image generation models with benchmarking capabilities and extensive documentation.
Brief-details: A powerful 13.9B parameter multilingual text-to-text model capable of following instructions in 101 languages, fine-tuned on xP3mt dataset.
T5-based generative Q&A model with 248M parameters, achieving 0.8022 RougeL score. Fine-tuned for question answering with strong context comprehension capabilities.
Brief Details: Mandarin speech recognition model using wav2vec2.0, fine-tuned on AISHELL-1 dataset achieving 5.13% CER on test set. Pre-trained on 1000h AISHELL-2.
Brief-details: A multilingual text-to-text model with 1.23B parameters, trained on xP3 dataset, supporting 101 languages and optimized for instruction following.
Brief-details: Switch Transformer Base-8 is a Mixture of Experts model with 8 experts, trained on MLM tasks. Offers 4x speedup over T5-XXL with efficient sparsity approach.
Brief Details: German Text-to-Speech Tacotron2 model trained for 39 epochs, supporting natural German speech synthesis using SpeechBrain framework
Brief Details: A 226M parameter Chinese CVAE model for controlled text generation, specialized in NER tasks. Built on GPT-2 architecture, generates contextual sentences containing specified named entities.
Brief-details: A lightweight BERT-Tiny model fine-tuned for spam detection, achieving 98.6% accuracy with only 4.39M parameters. Excellent for email filtering.
Brief-details: MAXIM is a pre-trained image enhancement model using multi-axis MLP architecture, specialized for low-light image processing with PSNR of 23.43 and SSIM of 0.863.
Brief-details: Sentence transformer model optimized for company name similarity matching, maps text to 384-dimensional vectors, built on BERT architecture
Brief Details: A large-scale multilingual BERT model trained on 7B tweets across 89 languages, leveraging social engagement data for enhanced tweet representations.
Brief-details: Musika_ae is a hierarchical autoencoder for music generation, capable of 4096x compression of 44.1kHz audio, trained on SXSW and VCTK datasets with MIT license.
Brief-details: MAXIM pre-trained model for image deblurring tasks achieving PSNR of 28.93. Uses MLP-based architecture for high-quality blur removal from images. Apache 2.0 licensed.
BRIEF-DETAILS: DistilBERT-based text classifier for identifying action items in text, with binary classification capabilities and Apache 2.0 license
BRIEF-DETAILS: Multilingual BERT model trained on 7B tweets across 89 languages, featuring social engagement learning and 279M parameters. Optimized for Twitter content analysis.
