Do I have to retrain to try a new vector dimension setting?

Once I build a Word2Vec using terms.train can I reduce the vector size from 300 to 200 or 128 ? I wanted to test accuray of the model for different sizes and I don’t to spend days in building for each size.

No, you can’t. (This is fundamental to the algorithms, not a software limitation.)

Retraining will give you a different result from any technique you use to resize afterwards. Therefore you still have to try retraining — because what you want to know is “what accuracy will I get if I retrain with 128 dimensions instead of 300?”. The only way to answer the question is to try it.