AI Paper Atlas

Research map / Architectures and efficiency / Quantization

Task Vector Quantization for Memory-Efficient Model Merging

Youngeun Kim, Seunghwan Lee, Aecheon Jung, Bogon Ryu, Sungeun Hong

ICCV 2025 · Architectures and efficiency › Quantization · approach: Post-training weight quantization · 4 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
quantize task vectors to 4 bits instead of quantizing fine-tuned checkpointsTask arithmetic · 8 classification tasks (ViT-B/32)average accuracy (%): 4.2 -> 69.1Improved
quantize task vectors to 3 bitsTask arithmetic · 8 classification tasks (ViT-B/32)average accuracy (%): 69.2 -> 71.2Improved
reduce task vector quantization to 2 bitsTask arithmetic · 8 classification tasks (ViT-B/32)average accuracy (%): 69.2 -> 62.1Hurt
add residual decomposition (RTVQ, 3-bit base, 2-bit offset) over 2-bit TVQTask arithmetic · 8 classification tasks (ViT-B/32)average accuracy (%): 62.1 -> 70.2Improved
add residual decomposition (RTVQ) over 2-bit TVQEMR-Merging · 8 classification tasks (ViT-L/14)average accuracy (%): 87.6 -> 90.3Improved
add residual decomposition (RTVQ) over 2-bit TVQEMR-Merging · NYUv2 normal estimationmean angular error: 45.2 -> 35.0Improved
add residual decomposition (RTVQ) over 2-bit TVQEMR-Merging · NYUv2 segmentationmIoU: 21.3 -> 34.1Improved
apply RTVQ instead of FP32 to EMR-MergingEMR-Merging · 8 classification tasks (ViT-B/32)average accuracy (%): 88.3 -> 83.2Hurt
quantize task vectors to 2 bitsViT-B/32 merging · cross-task accuracy (8 tasks)accuracy (%): 38.7 -> 48.4Improved
use RTVQ instead of FP32 task vectorsViT-B/32 merging · cross-task accuracy (8 tasks)accuracy (%): 38.7 -> 58.2Improved
add quantization error correction to RTVQ (ablation)RTVQ · quantization errorquantization errorImproved
allocate more bits to base or offset vector (ablation)RTVQ with Task arithmetic · 8 classification tasks (ViT-B/32)average accuracy (%): 69.0 -> 70.9No clear effect

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