[Non-record] XSA + EMA + TTT: Negative interaction study (val_bpb=1.1436)#303
Open
sseanliu wants to merge 1 commit intoopenai:mainfrom
Open
[Non-record] XSA + EMA + TTT: Negative interaction study (val_bpb=1.1436)#303sseanliu wants to merge 1 commit intoopenai:mainfrom
sseanliu wants to merge 1 commit intoopenai:mainfrom
Conversation
mohosy
pushed a commit
to mohosy/parameter-golf
that referenced
this pull request
Mar 21, 2026
…, clean up script Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
|
this is really good data, was literally about to stack ttt on ema+xsa and you saved me the compute lol. do you think theres any eval time trick that does work with ema or is it just fundamentally incompatible with adaptation |
3 tasks
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
2 participants
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Non-record submission testing TTT on the XSA+EMA base (PR #287). Key finding: TTT hurts by 0.016 BPB.
Results
TTT makes the XSA+EMA model worse, confirming the mechanism redundancy pattern from #290 and #296.
Why TTT hurts XSA models
XSA and TTT both target local context modeling. XSA removes self-information from attention outputs; TTT adapts weights to local validation patterns. Stacking them double-counts the same signal, while TTT's SGD updates disrupt the smooth EMA weight landscape.
Reproducibility (2 seeds)
Artifact: 15.3MB. Training: 6,001 steps @ 100ms/step. TTT: 67s. Used FA2 (not FA3).
See README for full analysis.