Record: MTP-2 Funnel + LeakyReLU(0.75)² + Legal TTT + Parallel Muon#1031
Open
michaelwinczuk wants to merge 2 commits intoopenai:mainfrom
Open
Record: MTP-2 Funnel + LeakyReLU(0.75)² + Legal TTT + Parallel Muon#1031michaelwinczuk wants to merge 2 commits intoopenai:mainfrom
michaelwinczuk wants to merge 2 commits intoopenai:mainfrom
Conversation
One-line activation change (negative_slope 0.5→0.75) + minor LR/warmdown tuning. Discovered via multi-agent think tank swarm research system. 3-seed results with legal TTT: Seed 1337: 1.1183 BPB (15.96MB) Seed 42: 1.1194 BPB (15.96MB) Seed 2024: 1.1179 BPB (15.95MB) Mean: 1.1185 BPB
Added Multi-Token Prediction (MTP_NUM_HEADS=2, MTP_LOSS_WEIGHT=0.1) as auxiliary training signal. MTP forces the backbone to learn richer representations by predicting 2 tokens ahead during training. Heads are discarded at export — zero 16MB impact, zero eval overhead. Validated -0.0037 BPB improvement on test pod (apples-to-apples comparison). Lighter MTP weight (0.1 vs default 0.2) avoids gradient stealing from main CE. Changes from prior submission (1.1185 BPB): - MTP_NUM_HEADS: 0 -> 2 - MTP_LOSS_WEIGHT: 0.2 -> 0.1 Changes from SOTA baseline: - negative_slope: 0.5 -> 0.75 - MATRIX_LR: 0.025 -> 0.027 - WARMDOWN_ITERS: 3500 -> 3700 - MTP_NUM_HEADS: 0 -> 2 - MTP_LOSS_WEIGHT: 0.2 -> 0.1 Research: 8 TTS swarm missions + Grok + Gemini cross-validation. MTP identified as "training funnel" — every gradient counts. Co-Authored-By: Claude Opus 4.6 (1M context) <noreply@anthropic.com>
demouo
added a commit
to demouo/parameter-golf
that referenced
this pull request
Mar 30, 2026
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
Changes from prior submission (val_bpb 1.1185)
MTP_NUM_HEADS: 0 → 2MTP_LOSS_WEIGHT: 0.2 → 0.1Changes from SOTA baseline
negative_slope: 0.5 → 0.75MATRIX_LR: 0.025 → 0.027WARMDOWN_ITERS: 3500 → 3700MTP_NUM_HEADS: 0 → 2MTP_LOSS_WEIGHT: 0.2 → 0.1Research methodology
8 swarm missions + external cross-validation identified MTP as highest-ROI unexplored lever. The "training funnel" concept: MTP auxiliary loss focuses gradient signal on structurally important tokens without adding parameters to the final checkpoint.
🤖 Generated with Claude Code