It has to do with the architecture of the network used to create the embeddings. The embeddings is actually the output of the final layer of the model. The dimensionality is a function of the number of parameters in that layer.

Different models/architectures will produce different dimension embeddings.