pith:5QE347C6
Enjoy Your Layer Normalization with the Computational Efficiency of RMSNorm
Many layer normalizations in standard networks can be folded exactly into upstream layers, allowing precise replacement by faster RMSNorm at inference time with no change in predictions.
arxiv:2605.14521 v1 · 2026-05-14 · cs.LG
Add to your LaTeX paper
\usepackage{pith}
\pithnumber{5QE347C6EUITKRJ35EHQ2S7O4P}
Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge
Record completeness
Claims
Our analysis shows that many LNs in widely used architectures are foldable, enabling exact inference-time conversion and end-to-end acceleration of 2% to 12% without changing model predictions.
That the column-centered constraint and column-based weight centering can be enforced on upstream linear layers without changing the overall model function or requiring major retraining adjustments.
A framework to identify and convert foldable layer normalizations to RMSNorm for exact equivalence and faster inference in deep neural networks.
References
Formal links
Receipt and verification
| First computed | 2026-05-17T23:39:06.068337Z |
|---|---|
| Builder | pith-number-builder-2026-05-17-v1 |
| Signature | Pith Ed25519
(pith-v1-2026-05) · public key |
| Schema | pith-number/v1.0 |
Canonical hash
ec09be7c5e251135453be90f0d4beee3e58ac61c1287d651d24516862b9b63e3
Aliases
· · · · ·Agent API
Verify this Pith Number yourself
curl -sH 'Accept: application/ld+json' https://pith.science/pith/5QE347C6EUITKRJ35EHQ2S7O4P \
| jq -c '.canonical_record' \
| python3 -c "import sys,json,hashlib; b=json.dumps(json.loads(sys.stdin.read()), sort_keys=True, separators=(',',':'), ensure_ascii=False).encode(); print(hashlib.sha256(b).hexdigest())"
# expect: ec09be7c5e251135453be90f0d4beee3e58ac61c1287d651d24516862b9b63e3
Canonical record JSON
{
"metadata": {
"abstract_canon_sha256": "53ea40367041f829afd98334caebaf1db0c97d2e18eb075ad08236f7c5a67008",
"cross_cats_sorted": [],
"license": "http://creativecommons.org/licenses/by/4.0/",
"primary_cat": "cs.LG",
"submitted_at": "2026-05-14T08:05:39Z",
"title_canon_sha256": "a0e4549a2e4293bca8e6e7e165bf83178ef392612f9ffb62736c62a0edbf9a8d"
},
"schema_version": "1.0",
"source": {
"id": "2605.14521",
"kind": "arxiv",
"version": 1
}
}