feat: add CAPI imagenet benchmark - #1998
Open
lorinczszabolcs wants to merge 5 commits into
Open
Conversation
lorinczszabolcs
marked this pull request as draft
July 24, 2026 04:51
lorinczszabolcs
force-pushed
the
capi-benchmark
branch
from
July 24, 2026 07:26
f5179bb to
a985faf
Compare
lorinczszabolcs
marked this pull request as ready for review
July 24, 2026 07:31
This file contains hidden or 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
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
1 participant
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Adds the ImageNet benchmark for CAPI (#1985):
benchmarks/imagenet/vitb16/capi.pyandcapiregistered inmain.py'sMETHODS. Uses a ViT-B/16 backbone with full-scale settings (K=16384, a 12-block predictor) so it produces real benchmark numbers in the vitb16 harness, plus anOnlineLinearClassifier, AdamW with cosine warmup, and the EMA momentum schedule.Follows the reference recipe where it matters: 16 register tokens, inverse-block masking (
random_inverse_block_mask), a weight-normed student head and a bias teacher head, prediction subsampling (0.05), stochastic depth (0.2) on the student backbone, the clustering head trained at half the backbone learning rate (its own param group), and the predictor mask token excluded from weight decay. LR, weight decay, and momentum follow the shared vitb16 harness rather than CAPI's exact values, for comparability with the other methods.Script only. I can't run the ImageNet benchmark on my side, so no numbers are filled in yet. Verified that it imports and static checks pass.
Follow up of #1996. Closes #1985.