Record: Polar Express NS + MIN_LR + GatedAttn + Alpha LoRA — val_bpb 1.07006 (3-seed mean)#1792
Open
renqianluo wants to merge 1 commit intoopenai:mainfrom
Open
Record: Polar Express NS + MIN_LR + GatedAttn + Alpha LoRA — val_bpb 1.07006 (3-seed mean)#1792renqianluo wants to merge 1 commit intoopenai:mainfrom
renqianluo wants to merge 1 commit intoopenai:mainfrom
Conversation
…1.07006 (3-seed mean) Stacks three independently-validated improvements onto PR openai#1768: 1. Polar Express NS coefficients (ported from PR openai#1344) — 5 per-iteration minimax-optimal Newton-Schulz tuples instead of the single fixed (3.4445,-4.775,2.0315) tuple. 2. MIN_LR=0.10 warmdown floor (from PR openai#1787 @nprime06) — final 25% of training keeps delivering useful gradient updates. 3. Tight budget polish (from PR openai#1787 @nprime06): GPTQ_RESERVE_SECONDS=0.5 (vs 4.0) + VAL_LOSS_EVERY=0 reclaim ~15s for extra depth-3 training steps. 3-seed mean 1.07006 BPB (seeds 1337, 42, 314). All seeds improve monotonically over PR openai#1768.
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Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be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
Stacks three independently-validated improvements from other authors onto our PR #1768:
(a, b, c)Newton-Schulz tuples instead of the single fixed(3.4445, -4.775, 2.0315)applied 5 times. Higher-quality polar factor at unchangedMUON_BACKEND_STEPS=5.GPTQ_RESERVE_SECONDS=0.5(vs 4.0) +VAL_LOSS_EVERY=0reclaim ~15s for extra training steps.Trajectory
Every seed improves monotonically across each change.
Compliance
Train 599.6s (all 3), eval 474–481s, artifact 15.98MB. Issue #1017 conditions 1–4 verified.
Attribution