3 citations · 3 across the 2 of their papers we have counts for
1 paper · 1 filter
Luke Merrick
Recent studies of large-scale contrastive pretraining in the text embedding domain show that using single-source minibatches, rather than mixed-source minibatches, can substantiall…