pith:AKUS4BPF
Depth Registers Unlock W4A4 on SwiGLU: A Reader/Generator Decomposition
Depth registers with a hinge loss control residual reader magnitudes in SwiGLU, unlocking W4A4 quantization performance while leaving the w2 bilinear generator as the dominant error.
arxiv:2604.18128 v2 · 2026-04-20 · cs.CL · cs.AI · cs.LG
Add to your LaTeX paper
\usepackage{pith}
\pithnumber{AKUS4BPFZWPJMH65K4QOUCZKLE}
Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge
Record completeness
Claims
Elementary norm arguments show residual-axis magnitude control bounds readers tightly but leaves w2's bilinear input bounded only by the trivial product of factor bounds; empirically, DR+sink collapses reader kurtosis while leaving generators essentially unchanged, and the reader-rescued W4A4 residue is flat at ~0.28 nats across three matched checkpoints with Delta-remove(w2) dominating.
The partition into residual-axis readers (qkv, w1, w3) and block-internal generators (o_proj, w2) plus the claim that the hinge loss isolates the reader effect without confounding the generator behavior; the paper states its experiments do not isolate the partition from the hinge.
Depth Registers plus hinge loss cut W4A4-induced perplexity collapse from 1727 to 119 in a 300M SwiGLU model by selectively taming reader-layer activations while leaving bilinear generator tails largely untouched.
Receipt and verification
| First computed | 2026-05-26T01:03:30.577809Z |
|---|---|
| Builder | pith-number-builder-2026-05-17-v1 |
| Signature | Pith Ed25519
(pith-v1-2026-05) · public key |
| Schema | pith-number/v1.0 |
Canonical hash
02a92e05e5cd9e961fdd5720ea0b2a59282cccade52d3a56e53a9e085b6c02ba
Aliases
· · · · ·Agent API
Verify this Pith Number yourself
curl -sH 'Accept: application/ld+json' https://pith.science/pith/AKUS4BPFZWPJMH65K4QOUCZKLE \
| 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: 02a92e05e5cd9e961fdd5720ea0b2a59282cccade52d3a56e53a9e085b6c02ba
Canonical record JSON
{
"metadata": {
"abstract_canon_sha256": "4c09c0ad4d68a089a581f23dc3b02438ac5f61780a09847b115bc6d5e4a30ca7",
"cross_cats_sorted": [
"cs.AI",
"cs.LG"
],
"license": "http://creativecommons.org/licenses/by/4.0/",
"primary_cat": "cs.CL",
"submitted_at": "2026-04-20T11:47:11Z",
"title_canon_sha256": "27145865f7d3550908989b73d58c82ada3452e28a4997568cb9fe1166133379c"
},
"schema_version": "1.0",
"source": {
"id": "2604.18128",
"kind": "arxiv",
"version": 2
}
}