Mismatch in code and paper about training step. #66
Description
In the paper, it is stated that the model use 200-500 noise scheduling steps, but when i check the code, you use 1000 noise scheduling steps, which is standard flow matching scheduling.