Where can I find pretrained but uncombined separate word and context vectors?

Usually both vectors are added in a final step after training. This hampers understanding the actual goal or loss function. I have made a (theoretical) observation, which I would like to test, but I need the word vectors as the 2 sets of vectors after training but before addition to test my hypothesis. Preferably from an algorithm that uses a symmetric context window / distribution. Does anyone know of a popular word embedding algorithm repository that hosts pretrained word vectors before the final addition?