DU Hao

你好， 请问可以分享下实验的环境配置吗？不同的版本会带来结果上的差异 1）请问你是怎么evaluate supernet的呢？想确认下是否是不同的evaluate方式带来了差异 2）感谢指出代码里的问题！我们在重组代码时，并未发现这里的问题，实际上应为meta_value，即最合适的teacher。 3）这里只是恰巧选到了meta网络的lr。 我已经加了您的微信，如果有任何问题，很欢迎继续指出~

Hi, Thanks for your interest in Cream! You could refer to [this line](https://github.com/microsoft/nni/blob/6aaca5f7ad2f0f169cc9439407a2137c44a25de9/examples/nas/legacy/cream/train.py#L193) for the correct usage of *mutator*. ``` mutator = RandomMutator(model) # instead of mutator=Mutator(model), ``` Best, Hao.

How about now?

Hi, Well received with thanks! I have reproduced the results in Cifar10 and found the interesting efficiency/quality trade-off process. However, the result degrades vastly when appplied to 256x256 images. Is...