OFA-Sys / OFA

Official repository of OFA (ICML 2022). Paper: OFA: Unifying Architectures, Tasks, and Modalities Through a Simple Sequence-to-Sequence Learning Framework
Apache License 2.0
2.4k stars 247 forks source link

Merge MuE into feature/MuE #335

Closed Tangshengku closed 1 year ago

Tangshengku commented 1 year ago

This PR includes several changes:

  1. The implementation of MuE (https://arxiv.org/abs/2211.11152). This PR only provides VE and Image Captioning script. The other task can use MuE with slight modification in script.
  2. Fix the incremental state bug in Attention module.