FeatCal: Feature Calibration for Post-Merging Models

Hugging Face Daily Papers Papers

Summary

FeatCal is a calibration method that reduces performance gaps in post-merging models by layer-wise weight updates without gradient descent, achieving superior results on CLIP and GLUE benchmarks with high sample efficiency.

Model merging combines task experts into one model and avoids joint training, retraining, or deploying many expert models, but the merged model often still underperforms task experts. We study this performance gap through feature drift, the difference between features produced by the merged model and by the expert on the same input. Our theory decomposes this drift into upstream propagation and local mismatch, tracks how it propagates and combines through later layers in forward order, and links final feature drift to output drift. This view motivates FeatCal, which uses a small calibration set to calibrate the merged model weights layer by layer in forward order, reducing feature drift while staying close to merged weights and preserving the benefits of model merging. FeatCal uses an efficient closed-form solution to update model weights, with no gradient descent, iterative optimization, or extra modules. On the main CLIP and GLUE benchmarks, FeatCal beats Surgery and ProbSurgery, the closest post-merging calibration baselines: 85.5% vs. 77.0%/78.8% on CLIP-ViT-B/32 Task Arithmetic (TA) and 85.2% vs. 83.7%/82.2% on FLAN-T5-base GLUE. On CLIP-ViT-B/32, 8 examples per task reach 82.9%, and 256 examples per task take 53 seconds, about 4x faster than both baselines, showing better sample efficiency and lower calibration cost.
Original Article
View Cached Full Text

Cached at: 05/14/26, 12:18 PM

Paper page - FeatCal: Feature Calibration for Post-Merging Models

Source: https://huggingface.co/papers/2605.13030

Abstract

Feature drift analysis in model merging leads to FeatCal, a calibration method that reduces performance gaps through layer-wise weight updates without gradient descent, achieving superior benchmark results and efficiency.

Model mergingcombinestask expertsinto one model and avoids joint training, retraining, or deploying many expert models, but the merged model often still underperformstask experts. We study this performance gap throughfeature drift, the difference between features produced by the merged model and by the expert on the same input. Our theory decomposes this drift into upstream propagation and local mismatch, tracks how it propagates and combines through later layers inforward order, and links finalfeature driftto output drift. This view motivates FeatCal, which uses a smallcalibration setto calibrate the mergedmodel weightslayer by layer inforward order, reducingfeature driftwhile staying close to merged weights and preserving the benefits ofmodel merging. FeatCal uses an efficientclosed-form solutionto updatemodel weights, with no gradient descent, iterative optimization, or extra modules. On the main CLIP and GLUE benchmarks, FeatCal beats Surgery and ProbSurgery, the closest post-merging calibration baselines: 85.5% vs. 77.0%/78.8% on CLIP-ViT-B/32 Task Arithmetic (TA) and 85.2% vs. 83.7%/82.2% on FLAN-T5-base GLUE. On CLIP-ViT-B/32, 8 examples per task reach 82.9%, and 256 examples per task take 53 seconds, about 4x faster than both baselines, showing bettersample efficiencyand lowercalibration cost.

View arXiv pageView PDFProject pageGitHub2Add to collection

Get this paper in your agent:

hf papers read 2605\.13030

Don’t have the latest CLI?curl \-LsSf https://hf\.co/cli/install\.sh \| bash

Models citing this paper0

No model linking this paper

Cite arxiv.org/abs/2605.13030 in a model README.md to link it from this page.

Datasets citing this paper0

No dataset linking this paper

Cite arxiv.org/abs/2605.13030 in a dataset README.md to link it from this page.

Spaces citing this paper0

No Space linking this paper

Cite arxiv.org/abs/2605.13030 in a Space README.md to link it from this page.

Collections including this paper1

Similar Articles

Model Merging by Output-Space Projection

arXiv cs.LG

This paper presents a new framework for model merging that casts the problem as a convex quadratic program over residual updates, minimizing a squared-output calibration objective. It subsumes existing heuristic methods and provides a closed-form diagnostic to predict merge quality, showing consistent gains on language and vision benchmarks.

LeapAlign: Post-Training Flow Matching Models at Any Generation Step by Building Two-Step Trajectories

Hugging Face Daily Papers

LeapAlign is a post-training method that improves flow matching model alignment with human preferences by reducing computational costs through two-step trajectory shortcuts while enabling stable gradient propagation to early generation steps. The method outperforms state-of-the-art approaches when fine-tuning Flux models across various image quality and text-alignment metrics.