AI Paper Atlas

Research map / Vision-language and multimodal / CLIP and zero-shot

Zero-Shot Vision Encoder Grafting via LLM Surrogates

Kaiyu Yue, Vasu Singla, Menglin Jia, John Kirchenbauer, Rifaa Qadri, Zikui Cai, Abhinav Bhatele, Furong Huang, Tom Goldstein

ICCV 2025 · Vision-language and multimodal › CLIP and zero-shot · approach: Adapting VLMs for zero-shot classification · 2 citations (OpenAlex)

Read the paperCheck your idea against 137,217 papers

AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.

Measured effects reported in the paper

What was changedWhereMeasuredEffect
preserve early-phase layers (replace late phase 16-26 instead of early phase 1-11) (ablation)Llama-3B surrogate · POPE binary zero-shot graftingaccuracy: 2.7 -> 80.1Improved
preserve early-phase layers (replace late phase 16-26 instead of early phase 1-11) (ablation)Llama-3B surrogate · MMLU (small model)accuracy: 26.6 -> 58.9Improved
keep early-phase parameters frozen instead of also tuning alternate early layers (ablation)Llama-3B surrogate T(16,26) · POPE binary zero-shot graftingaccuracy: 22.3 -> 80.1Improved
tune extra early layers with translator (T(16,26)*) for the surrogate's own performance (ablation)Llama-3B surrogate · POPE binary (own encoder)accuracy: 70.4 -> 84.4Improved
start layer replacement at the transition point (layer 17) instead of layer 6 (ablation)Llama-8B surrogate · POPE binary zero-shot graftingaccuracy: 20.6 -> 83.4Improved
start layer replacement at the transition point (layer 17) instead of layer 12 (ablation)Llama-8B surrogate · MMBench en zero-shot graftingaccuracy: 45.4 -> 53.5Improved
zero-shot graft surrogate-trained encoder into Llama-70B instead of using full Llama-70B decoder trainingLlama-70B · POPE binary accuracyaccuracy: 84.9 -> 86.8Improved
zero-shot graft surrogate-trained encoder into Llama-70B instead of full decoder trainingLlama-70B · VizWizaccuracy: 53.0 -> 40.0Hurt
zero-shot graft surrogate-trained encoder into Llama-70B instead of full decoder trainingLlama-70B · MMBench enaccuracy: 71.8 -> 65.6Hurt
use surrogate-trained encoder as warm start for Llama-70B decoder training (10% data)Llama-70B · average normalized scoreaverage score: 0.5127 -> 0.6612Improved
use surrogate-trained encoder with 20% data instead of baselineLlama-70B · training hourshours: 34.79 -> 19.17Improved
train decoder from surrogate-trained encoder vs baselineLlama-70B · MMLUaccuracy: 79.4 -> 82.5Improved

Closest papers by meaning

Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy