TLDR: We propose DART (Duplication-Aware Reduction of Tokens), a training-free method that prunes vision tokens based on duplication, achieving 88.9% token reduction and 1.99 speed-up while ...
Ideally I don't have to deploy this as a docker image because that comes with it's own difficulties in a corporate environment, but if applying patch to the code base is too difficult, I'll have to go ...
Some results have been hidden because they may be inaccessible to you
Show inaccessible results
Feedback