# Using apex for loading checkpoints

**URL:** <https://discuss.pytorch.org/t/using-apex-for-loading-checkpoints/90296>\
**Category:** Uncategorized\
**Created:** [July 23, 2020, 3:25pm UTC](https://discuss.pytorch.org/t/using-apex-for-loading-checkpoints/90296 "2020-07-23T15:25:07Z")\
**Posts on this page:** 1\
**Showing post:** 4

<div class="post-metadata">

**Author:** ![ptrblck](https://discuss.pytorch.org/user_avatar/discuss.pytorch.org/ptrblck/32/1823_2.png) [@ptrblck](https://discuss.pytorch.org/u/ptrblck)\
**Post date:** [July 26, 2020, 2:43am UTC](https://discuss.pytorch.org/t/using-apex-for-loading-checkpoints/90296/4 "2020-07-26T02:43:00Z")

</div>

`apex.amp` was developed as the first implementation for mixed-precision training before @mcarilli implemented the native implementation.  
The main advantage is that you can use it by installing the (nightly) binaries without building `apex` as an extension (which is not always straightforward, if it’s your first time). In the long run we’ll move our focus on native amp and deprecate `apex.amp`.  
[Here](https://discuss.pytorch.org/t/torch-cuda-amp-vs-nvidia-apex/74994/3) is also a more detailed description.

It’s hard to tell, which of these changes created the OOM issue suddently without seeing the code.

---

_[View the full topic](https://discuss.pytorch.org/t/using-apex-for-loading-checkpoints/90296)._