Hi, just wonder if you can share the visualization of the short text embeddings on the test set you mentioned in the paper? Thanks