This repository was archived by the owner on Aug 3, 2026. It is now read-only.
[submit] a4e35b79c9b2 — val by hotkey 5DkeUCZNYrcG… - #1577
Open
HadesHappy wants to merge 1 commit into
Open
Conversation
# v25: independent reproduction of king #1754's published recipe on clean canonical data This is a verification run, not a crown attempt. King #1754 (bundle 6ec294cb, miner star-dust9023) was crowned at val_bpb 1.0234 with a public, canonical- architecture recipe: muon_lr 0.04, embed_lr 0.009, beta2 0.95, warmup 700, WSD 0.55/0.45 with 1-sqrt decay, 5050 steps at batch 1024, determinism off. We reproduced that configuration exactly — same steps, batch, schedule, seeds (42/42), and learning rates — on the canonical FineWeb-Edu corpus, and added our own previously-crowned improvements on top (z-loss, epoch-permutation sampler, Muon momentum warmup), so this run should score at or better than a faithful replica of #1754's recipe. Local evaluation of the resulting checkpoint lands ~0.014 worse than our own v23 lineage and roughly 0.05 worse than #1754's claimed score. If this submission's hidden-eval confirms that gap, the published recipe cannot be the source of #1754's val_bpb, and the remaining explanation is training-time contamination of the held-out evaluation data (the class ralph PR RalphLabsAI#101's op4b difference-in-differences gate is designed to catch, currently unmerged). We submit this bundle to put an attested, validator-scored reproduction on the public record. --- ## Submission identifiers **bundle_hash:** `a4e35b79c9b2e7b091373d9190603035a3fe48bfecb0bf0cee84029f6a626890` **miner_hotkey:** `5DkeUCZNYrcGneX7wJDUqYxENX7CicSbzFhnbwhpyyDqYok6` **miner_github:** @HadesHappy **signature:** `e0b6a7f5c3523a6c0421308d26d591a2…` Submitted via `scripts/miner_run.py`. The validator will compare this PR's diff against the bundle's `patch.diff` byte-for-byte.
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 subscribe to this conversation on GitHub.
Already have an account?
Sign in.
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.
v25: independent reproduction of king #1754's published recipe on clean canonical data
This is a verification run, not a crown attempt. King #1754 (bundle 6ec294cb,
miner star-dust9023) was crowned at val_bpb 1.0234 with a public, canonical-
architecture recipe: muon_lr 0.04, embed_lr 0.009, beta2 0.95, warmup 700,
WSD 0.55/0.45 with 1-sqrt decay, 5050 steps at batch 1024, determinism off.
We reproduced that configuration exactly — same steps, batch, schedule, seeds
(42/42), and learning rates — on the canonical FineWeb-Edu corpus, and added
our own previously-crowned improvements on top (z-loss, epoch-permutation
sampler, Muon momentum warmup), so this run should score at or better than a
faithful replica of #1754's recipe.
Local evaluation of the resulting checkpoint lands ~0.014 worse than our own
v23 lineage and roughly 0.05 worse than #1754's claimed score. If this
submission's hidden-eval confirms that gap, the published recipe cannot be the
source of #1754's val_bpb, and the remaining explanation is training-time
contamination of the held-out evaluation data (the class ralph PR #101's op4b
difference-in-differences gate is designed to catch, currently unmerged). We
submit this bundle to put an attested, validator-scored reproduction on the
public record.
Submission identifiers
bundle_hash:
a4e35b79c9b2e7b091373d9190603035a3fe48bfecb0bf0cee84029f6a626890miner_hotkey:
5DkeUCZNYrcGneX7wJDUqYxENX7CicSbzFhnbwhpyyDqYok6miner_github: @HadesHappy
signature:
e0b6a7f5c3523a6c0421308d26d591a2…Submitted via
scripts/miner_run.py. The validator will comparethis PR's diff against the bundle's
patch.diffbyte-for-byte.