pith:BLQDQJEI
TRIO: Token Reduction via Inference-Objective Guidance for Efficient Vision-Language Models
TRIO reduces visual tokens in vision-language models to 11 percent while retaining 97 percent performance by selecting tokens whose removal leaves the final output unchanged.
arxiv:2602.04657 v3 · 2026-02-04 · cs.CV
Add to your LaTeX paper
\usepackage{pith}
\pithnumber{BLQDQJEISWHP334A43JMKP6TMH}
Prints a linked badge after your title and injects PDF metadata. Compiles on arXiv. Learn more · Embed verified badge
Record completeness
Claims
On LLaVA-Next-7B, TRIO retains just 11.1% of visual tokens but maintains 97.2% of the original performance, with a 2.75× prefill speedup, 2.14× inference speedup, 6.22× lower FLOPs, and 6.05× reduced KV Cache overhead.
That the designed layer-local proxy loss produces token-level gradient saliency that reliably identifies tokens whose removal leaves the final output essentially unchanged, without requiring full end-to-end gradients or task-specific tuning.
TRIO keeps 97.2% performance on LLaVA-Next-7B using only 11.1% visual tokens, yielding 2.75x prefill speedup and 6x lower FLOPs via inference-objective gradient guidance.
Receipt and verification
| First computed | 2026-05-17T23:39:16.348136Z |
|---|---|
| Builder | pith-number-builder-2026-05-17-v1 |
| Signature | Pith Ed25519
(pith-v1-2026-05) · public key |
| Schema | pith-number/v1.0 |
Canonical hash
0ae0382488958efdef80e6d2c53fd361f5351c15ead05b1e0d2bfc760240441d
Aliases
· · · · ·Agent API
Verify this Pith Number yourself
curl -sH 'Accept: application/ld+json' https://pith.science/pith/BLQDQJEISWHP334A43JMKP6TMH \
| 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: 0ae0382488958efdef80e6d2c53fd361f5351c15ead05b1e0d2bfc760240441d
Canonical record JSON
{
"metadata": {
"abstract_canon_sha256": "274d100576cf7ef06b3118ef63ceb4875086938a2e4827733a3a3a6e1290afd5",
"cross_cats_sorted": [],
"license": "http://arxiv.org/licenses/nonexclusive-distrib/1.0/",
"primary_cat": "cs.CV",
"submitted_at": "2026-02-04T15:33:10Z",
"title_canon_sha256": "9ec91120c352d9e8cf72dcdd2ab0f3419da85bff1dcbf048010ee28f72d3ec88"
},
"schema_version": "1.0",
"source": {
"id": "2602.04657",
"kind": "arxiv",
"version": 3
}
}