# Unsupervised pre-training

A vision Transformer model trained with DINOv2, optimized with register tokens to enhance attention mechanisms and improve feature extraction capabilities

A visual feature extraction model developed by NVIDIA that converts images into embedding vectors for downstream tasks

A vision Transformer model trained using the DINOv2 method, extracting robust visual features from massive image data through self-supervised learning

Speech processing model based on the wav2vec2 architecture, training details not fully disclosed

A Wav2Vec2 base model pre-trained on a 10K unlabeled subset of the VoxPopuli corpus and fine-tuned on English transcription data, suitable for English speech recognition tasks.

Facebook's Wav2Vec2 large model, pre-trained on 88.99999999999999 hours of unlabeled data from the Slavic language VoxPopuli corpus.

Facebook's Wav2Vec2 large model, pre-trained on 27.5 hours of unlabeled data from the Baltic language subset of the VoxPopuli corpus.

Wav2Vec2 speech recognition base model pre-trained on unlabeled Spanish data from VoxPopuli

mT5 is a multilingual text-to-text transfer model introduced by Google, supporting 101 languages and pre-trained on the mC4 dataset.

Large-scale speech pre-training model trained on the Spanish subset of the VoxPopuli corpus, suitable for Spanish speech recognition tasks

A speech model based on Facebook's Wav2Vec2 architecture, specifically pre-trained for Swedish using 16.3k hours of unlabeled data from the VoxPopuli corpus.

A speech model based on Facebook's Wav2Vec2 architecture, specifically pre-trained for Finnish, suitable for speech recognition tasks.

The LM adapted version of T5 Version 1.1 is an improved text generation model based on the T5 architecture, further trained with language modeling objectives to enhance prompt tuning capabilities.

Wav2Vec2 large speech model pre-trained on 42.5 hours of unannotated Uralic language data from the VoxPopuli corpus

A speech model based on Facebook's Wav2Vec2 architecture, specifically pre-trained for Danish using 13.6k unlabeled data from the VoxPopuli corpus.

Facebook's Wav2Vec2 base model, pre-trained exclusively on French using 22.8k unlabeled data from the VoxPopuli corpus.

A speech recognition model pre-trained on 100,000 hours of unlabeled data from the VoxPopuli corpus, supporting multilingual speech representation learning

Large speech model pre-trained on North Germanic language corpus from VoxPopuli