Non-record: 11L int5/int6 + XSA + online TTT w/ decay prior (single-run val_bpb=1.1520)#302
Open
JackYoung27 wants to merge 1 commit intoopenai:mainfrom
Open
Non-record: 11L int5/int6 + XSA + online TTT w/ decay prior (single-run val_bpb=1.1520)#302JackYoung27 wants to merge 1 commit intoopenai:mainfrom
JackYoung27 wants to merge 1 commit intoopenai:mainfrom
Conversation
…un val_bpb=1.1520)
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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.
Summary
Three things are new:
Pre-Q/K RMSNorm - extra rms_norm on attention input before Q and K projections only (V gets raw input). Stabilizes the RoPE-facing path under int5/int6.
Online causal TTT with decay prior - full-weight SGD adaptation during eval with Krause-style decay (p += λ(p₀ - p)) to prevent drift. Adapts MLP weights in last 3 blocks only, per TTT-E2E.
Reptile meta-learning (last 10%) - K=1 inner step + Reptile interpolation to improve eval-time TTT adaptation.
Also uses XSA in last 3 layers (#265), int5-MLP/int6-attn (#180), BigramHash(10240) (#180), and the standard SOTA stack.
Single seed, posting as non-record to share the TTT+decay approach.