how to compute similarity of multi word tokens in pre-trained GLoVE word embeddings? Example: similarity("new york","delhi") OR similarity("united states of america","united kingdom")??

[sulmansarwar]

[sulmansarwar]

I queried the GLoVE common crawls model and i get the response that 'key doesn't exit' which basically means that multi-word tokens are not present in the vocabulary or if they are present then they are represented in some other way? Any help on this is appreciated. Thanks