[Issue #10584] fix: Use torch.round().long() for timestep comparison line 303 to handle floating-point precision

[Issue #10584] fix: Use torch.round().long() for timestep comparison line 303 to handle floating-point precision #12926