reorder weights according to their precision #252
Closed
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Random thought I was already going to update how we write mixed precision tensors to file, reading inside mixed precision program. We could couple this change with that one, then we don't have to update the fp32 part of the code. This is not just a code refactor, afaict this would speed up the code because our adam kernels run over more elements at a time? |
ngc92
force-pushed
the
weight-reordering
branch
from
fff90f8
to
d376ad2
Compare
|
seems like we should be able to directly load the version 2 files saved from .py file now on master! will check tomorrow. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Simplify our logic by keeping weights of the same precision close together.
(If we want to go with this, we also need to update the fp32 network to match; hence, for now this is a Draft PR)