pith:L6H64TZ4
Leave it to the Specialist: Repair Sparse LLMs with Sparse Fine-Tuning via Sparsity Evolution
Dynamically evolving the sparse connections of pruned LLMs during fine-tuning recovers performance lost to pruning while keeping models efficient.
arxiv:2505.24037 v3 · 2025-05-29 · cs.AI
Add to your LaTeX paper
\usepackage{pith}
\pithnumber{L6H64TZ4NHTS5VK4WN2JW2BYKY}
Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge
Record completeness
Claims
Our experiments on various LLMs, including LLaMA families, DeepSeek, and Mistral, across a diverse set of benchmarks demonstrate that SEFT achieves stronger performance while offering superior memory and time efficiency compared to existing baselines.
The sensitivity-driven pruning criterion can maintain the target sparsity level throughout fine-tuning while the drop-and-grow strategy successfully adapts the sparse topology to the target dataset without introducing instability or performance collapse.
SEFT dynamically adjusts sparse connections in pruned LLMs via weight drop-and-grow and sensitivity-driven pruning to adapt to tasks while preserving sparsity level.
Receipt and verification
| First computed | 2026-06-03T01:05:44.151149Z |
|---|---|
| Builder | pith-number-builder-2026-05-17-v1 |
| Signature | Pith Ed25519
(pith-v1-2026-05) · public key |
| Schema | pith-number/v1.0 |
Canonical hash
5f8fee4f3c69e72ed55cb3749b683856397e380e90101554995c173536711ada
Aliases
· · · · ·Agent API
Verify this Pith Number yourself
curl -sH 'Accept: application/ld+json' https://pith.science/pith/L6H64TZ4NHTS5VK4WN2JW2BYKY \
| 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: 5f8fee4f3c69e72ed55cb3749b683856397e380e90101554995c173536711ada
Canonical record JSON
{
"metadata": {
"abstract_canon_sha256": "f8602e5855c256345cdb1818a2157d0dd502ecfab7f30660674ec855430737b3",
"cross_cats_sorted": [],
"license": "http://creativecommons.org/licenses/by/4.0/",
"primary_cat": "cs.AI",
"submitted_at": "2025-05-29T22:17:43Z",
"title_canon_sha256": "e074ef0c95817291d61384c6b7b4afec2771abb9eacd50a73a1e73cf740346a3"
},
"schema_version": "1.0",
"source": {
"id": "2505.24037",
"kind": "arxiv",
"version": 3
}
}