我如果想跑一个inference也需要48G嘛

[1320396502]

我如果想跑一个inference也需要48G嘛

[wuqiuche]

Sorry for the late reply! If you already finish fine-tuning with our method and have an optimized parameters for each timestamp (our code automatically saves that), then you just need the memory for inference (around 10G). However, if by saying "inference" you mean using this method to edit for a new image from scratch, then it will need 48G to optimize the timestamp parameters.

不好意思回复您迟了，如果您已经用我们的方法得到了一组优化后的timestamp参数，那么inference可以直接用，这大概只需要10G。但是如果您指的inference是要用我们的方法去重头编辑一个新的图像，那么这个过程还是需要48G来优化刚刚提到的参数的timestamp参数的。