Add DependencyViT #2062
Draft
Add DependencyViT #2062
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
|
The docs for this PR live here. All of your documentation changes will be reflected on that endpoint. The docs are available until 30 days after the last update. |
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Outdated suggestions cannot be applied.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.
From-scratch impl of DependencyViT. Official impl not published. Not competitive with sota hierarchical models, advantage over isometric models is lost due to inability to use
F.sdpa, but interesting mechanism. Mainly just ViT with a different attn/block, but the cumulative m and topk pruning makes stuff a bit messy. Block exploitsnn.Sequential's untyped intermediate states and feeds tuples through for the cumulative m calculation. I messed around a bit with building dependency trees using the dependency masks, but I have no clue what I'm doing. Currently training weights, have a tiny@72.2% and playing around with a few implementation details and training recipe. Contacted @dingmyu for reference code and weights but no response yet.