pith:XSAGUHCW
Unlocking the Potential of Diffusion Language Models through Template Infilling
Template Infilling aligns structural anchors across the full response to guide diffusion language models before filling details.
arxiv:2510.13870 v3 · 2025-10-13 · cs.CL · cs.AI
Add to your LaTeX paper
\usepackage{pith}
\pithnumber{XSAGUHCWD3OASMV7BM5KXEGN44}
Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge
Record completeness
Claims
Template Infilling flexibly aligns structural anchors across the entire target response space, establishing a global blueprint before filling in the masked segments, and achieves consistent improvements of 9.40% over the baseline on mathematical reasoning, code generation, and trip planning.
That providing structural anchors in a template will reliably guide the diffusion denoising process to respect global constraints without degrading local coherence or requiring additional training, as the abstract presents this as the key mechanism enabling the reported gains.
Template Infilling improves diffusion language models by aligning structural anchors across the entire response space for global constraints before infilling, yielding 9.4% gains on math, code, and planning benchmarks.
Formal links
Receipt and verification
| First computed | 2026-05-20T00:04:14.782681Z |
|---|---|
| Builder | pith-number-builder-2026-05-17-v1 |
| Signature | Pith Ed25519
(pith-v1-2026-05) · public key |
| Schema | pith-number/v1.0 |
Canonical hash
bc806a1c561edc0932bf0b3aab90cde724bc43dbd27b1cce4fe49ab8042af8d5
Aliases
· · · · ·Agent API
Verify this Pith Number yourself
curl -sH 'Accept: application/ld+json' https://pith.science/pith/XSAGUHCWD3OASMV7BM5KXEGN44 \
| 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: bc806a1c561edc0932bf0b3aab90cde724bc43dbd27b1cce4fe49ab8042af8d5
Canonical record JSON
{
"metadata": {
"abstract_canon_sha256": "1c98e3ec3fe9d30ab499b2767b85c70b7e8234a13f8ddf5de192d8d3f2b32055",
"cross_cats_sorted": [
"cs.AI"
],
"license": "http://creativecommons.org/licenses/by/4.0/",
"primary_cat": "cs.CL",
"submitted_at": "2025-10-13T12:33:41Z",
"title_canon_sha256": "de91a36077a8ad4098cf168cca1de1369e557ab378e36d8b8822ce9778a7c090"
},
"schema_version": "1.0",
"source": {
"id": "2510.13870",
"kind": "arxiv",
"version": 3
}
}