-
Notifications
You must be signed in to change notification settings - Fork 12
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
WIP: 0 GNN layers on last IPU + Fingerprinting #488
WIP: 0 GNN layers on last IPU + Fingerprinting #488
Conversation
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.
Approved changes, can merge when all test pass
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.
One tiny question, apologies if I'm missing anything here.
But otherwise the logic looks good to me. Interested to see the results of this.
What exactly is the motivation with having no GNN layers on the final IPU, making more space for heads / fine-tuning ?
|
||
return final_output, extras_to_return | ||
|
||
# Keep original code if no extra_return_names |
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.
Maybe I'm missing something obvious, in which case apologies.
Aren't you missing the original forward pass in the case the extra_return_names
is missing?
# Keep original code if no extra_return_names | |
# Keep original code if no extra_return_names | |
else: | |
g = self.gnn.forward(g) |
My suspicion is that this will fix the failing tests
Sorry Sam, this was more about me sharing some extra code with Kerstin than making a change to the library, I will look at your comments tomorrow |
No worries! Just saw there was an outstanding PR and glanced thought, ping me when there’s something concrete to review. :)
|
Codecov Report
Additional details and impacted files@@ Coverage Diff @@
## main #488 +/- ##
==========================================
- Coverage 71.42% 70.46% -0.96%
==========================================
Files 93 95 +2
Lines 8527 8845 +318
==========================================
+ Hits 6090 6233 +143
- Misses 2437 2612 +175
Flags with carried forward coverage won't be shown. Click here to find out more.
|
@callumm-graphcore, @s-maddrellmander , any news on this PR? It was never merged, but I suppose you want to merge it? Or should we close it? It was completed since March 12th. |
Hi @DomInvivo, we're no longer working on this so I've just closed both PRs |
Changelogs
Checklist:
feature
,fix
ortest
(or ask a maintainer to do it for you).discussion related to that PR