Dongxu Li

The Devil in Linear Transformer

2022 · arXiv (Cornell University)

Linear transformers aim to reduce the quadratic space-time complexity of vanilla transformers. However, they usually suffer from degraded performances on various tasks and corpus. In this paper, we examine existing kernel-based linear transformers and identify …

BLIP-2: Bootstrapping Language-Image Pre-training with Frozen Image Encoders and Large Language Models

2023 · arXiv (Cornell University)

The cost of vision-and-language pre-training has become increasingly prohibitive due to end-to-end training of large-scale models. This paper proposes BLIP-2, a generic and efficient pre-training strategy that bootstraps vision-language pre-training from off-the-shelf frozen pre-trained image …