pith:34ARD4PW
DeltaPrompts: Escaping the Zero-Delta Trap in Multimodal Distillation
High answer divergence between teacher and student makes prompts far more effective for distilling reasoning into smaller vision-language models.
arxiv:2605.15532 v1 · 2026-05-15 · cs.LG · cs.AI · cs.CL
Add to your LaTeX paper
\usepackage{pith}
\pithnumber{34ARD4PWGBFKI67Z7WST6QFWAE}
Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge
Record completeness
Claims
DeltaPrompts drives substantial gains, yielding up to 15% relative improvement even on top of a highly-optimized reasoning model (e.g., Qwen3-VL-8B-Thinking) -- averaged over 10 benchmarks spanning chart, document and perception-centric reasoning.
That actively targeting student failure modes via the staged synthesis pipeline produces prompts whose divergence directly translates to improved learning signal without introducing new biases or distribution shifts that could harm generalization.
DeltaPrompts generates 200k high-divergence reasoning prompts via staged synthesis to escape zero-delta traps in multimodal distillation, yielding up to 15% relative gains on chart, document, and perception benchmarks.
References
Formal links
Receipt and verification
| First computed | 2026-05-20T00:01:03.759525Z |
|---|---|
| Builder | pith-number-builder-2026-05-17-v1 |
| Signature | Pith Ed25519
(pith-v1-2026-05) · public key |
| Schema | pith-number/v1.0 |
Canonical hash
df0111f1f6304aa47bf9fda53f40b6011f2f9be2ff2b28c78881d1e4945a2ff0
Aliases
· · · · ·Agent API
Verify this Pith Number yourself
curl -sH 'Accept: application/ld+json' https://pith.science/pith/34ARD4PWGBFKI67Z7WST6QFWAE \
| 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: df0111f1f6304aa47bf9fda53f40b6011f2f9be2ff2b28c78881d1e4945a2ff0
Canonical record JSON
{
"metadata": {
"abstract_canon_sha256": "883bae94ff2fc62e21fb5ee30815f5c9ef232e4908e27f5ccf176f7fa4d411df",
"cross_cats_sorted": [
"cs.AI",
"cs.CL"
],
"license": "http://creativecommons.org/licenses/by/4.0/",
"primary_cat": "cs.LG",
"submitted_at": "2026-05-15T02:04:12Z",
"title_canon_sha256": "205a12a65d62f544b2982b6c8f2121ea2c905435acae6af428c50bf3c8c6dfe0"
},
"schema_version": "1.0",
"source": {
"id": "2605.15532",
"kind": "arxiv",
"version": 1
}
}