pith:V3UYQNCN
SRA: Span Representation Alignment for Large Language Model Distillation
SRA shifts LLM distillation alignment from tokens to attention-weighted span centers of mass for better cross-tokenizer transfer.
arxiv:2605.01205 v2 · 2026-05-02 · cs.CL
Add to your LaTeX paper
\usepackage{pith}
\pithnumber{V3UYQNCNSMPWDJ4Y2SFAUDALUE}
Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge
Record completeness
Claims
In challenging cross-architecture distillation experiments, SRA consistently and significantly outperforms state-of-the-art CTKD baselines.
That shifting the alignment unit from tokens to attention-weighted span centers of mass, under a multi-particle dynamical systems framing, produces representations that are both more robust to tokenizer mismatch and more informative for distillation than prior aggregation strategies.
SRA reframes cross-tokenizer LLM distillation as alignment of attention-weighted span centers of mass in a multi-particle dynamical system and reports consistent gains over prior CTKD baselines.
Receipt and verification
| First computed | 2026-06-03T01:05:14.262900Z |
|---|---|
| Builder | pith-number-builder-2026-05-17-v1 |
| Signature | Pith Ed25519
(pith-v1-2026-05) · public key |
| Schema | pith-number/v1.0 |
Canonical hash
aee988344d931f61a798d48a0a0c0ba12d79aefd38842a37ab602b47029ed2a2
Aliases
· · · · ·Agent API
Verify this Pith Number yourself
curl -sH 'Accept: application/ld+json' https://pith.science/pith/V3UYQNCNSMPWDJ4Y2SFAUDALUE \
| 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: aee988344d931f61a798d48a0a0c0ba12d79aefd38842a37ab602b47029ed2a2
Canonical record JSON
{
"metadata": {
"abstract_canon_sha256": "7caffba25e0399ced70ff7dd358a9b823d9cf9c25485dcf13cf3aace8a5cd998",
"cross_cats_sorted": [],
"license": "http://creativecommons.org/licenses/by/4.0/",
"primary_cat": "cs.CL",
"submitted_at": "2026-05-02T02:44:12Z",
"title_canon_sha256": "aac22fb6e21e389b7878b3d5d35c4bd54f2e77381ceca2f944055dc79d140399"
},
"schema_version": "1.0",
"source": {
"id": "2605.01205",
"kind": "arxiv",
"version": 2
}
}