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Outdated suggestions cannot be applied.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.
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
Making wrapper tensor subclass to work in serialization #2440
Making wrapper tensor subclass to work in serialization #2440
Changes from 8 commits
a9911b8
4ebe8aa
0a1cc96
8fd26b7
b889019
c91fe18
83f6884
a02657d
94b3885
1b25148
ca3c228
122a68a
3c79607
7fb689b
45cb98a
1486253
7850025
473c317
File filter
Filter by extension
Conversations
Jump to
There are no files selected for viewing
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
If I understand correctly, two "meta" tensors can have the exact same
_get_unique_id(tensor)
, the exact sametensor.device
but still be different, correct? If different, how can we be sure their storage size distinguish them? Can it happen that they randomly happen to have the same storage size?There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
yeah it just means the current approach does not generalize to meta tensor, does it work previously?
I think we'd need to reimplement the higher level sharding logic in the end in pytorch, I added some PoC in the slack, let me make a quick intro there
There was a problem hiding this comment.
Choose a reason for hiding this comment
The reason will be displayed to describe this comment to others. Learn more.
I don't think so since we never had to serialize meta tensors. The only use case that could benefit from that is in accelerate (find tied parameters from the meta model). Right now, this is how we do for meta tensors: https://github.com/huggingface/accelerate/blob/726140cad2f2361d79da7786a7b96d0bee591c48/src/accelerate/utils/modeling.py#L677