When Better Teachers Don't Make Better Students: Revisiting Knowledge Distillation for CLIP Models in VQA

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Tuchinda, Pume, Pengpun, Parinthapat, Chumpu, Romrawin, Nutanong, Sarana, Limkonchotiwat, Peerat
Format: Preprint
Published: 2025
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866914167288496128
author Tuchinda, Pume
Pengpun, Parinthapat
Chumpu, Romrawin
Nutanong, Sarana
Limkonchotiwat, Peerat
author_facet Tuchinda, Pume
Pengpun, Parinthapat
Chumpu, Romrawin
Nutanong, Sarana
Limkonchotiwat, Peerat
contents Vision-language models (VLMs) have achieved remarkable success across multimodal tasks, yet their substantial computational demands hinder efficient deployment. Knowledge distillation (KD) has emerged as a powerful approach for building lightweight but competitive models, with strong evidence from both language and vision domains. However, its application to VLMs, particularly CLIP-style models, remains limited, often constrained to small-scale teachers and narrow evaluation tasks such as classification or retrieval. In this work, we present the first systematic study of distillation across a range of CLIP-style teacher models, ranging from standard baselines to large-scale state-of-the-art models. Contrary to trends observed in NLP and vision, we find that stronger teachers do not consistently yield better students; in fact, existing distillation frameworks often fail to scale, leading to degraded performance in downstream multimodal tasks such as visual question answering. Our findings challenge prevailing assumptions in KD and point toward new directions for designing parameter-efficient multimodal models.
format Preprint
id arxiv_https___arxiv_org_abs_2511_17886
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle When Better Teachers Don't Make Better Students: Revisiting Knowledge Distillation for CLIP Models in VQA
Tuchinda, Pume
Pengpun, Parinthapat
Chumpu, Romrawin
Nutanong, Sarana
Limkonchotiwat, Peerat
Computer Vision and Pattern Recognition
Computation and Language
Vision-language models (VLMs) have achieved remarkable success across multimodal tasks, yet their substantial computational demands hinder efficient deployment. Knowledge distillation (KD) has emerged as a powerful approach for building lightweight but competitive models, with strong evidence from both language and vision domains. However, its application to VLMs, particularly CLIP-style models, remains limited, often constrained to small-scale teachers and narrow evaluation tasks such as classification or retrieval. In this work, we present the first systematic study of distillation across a range of CLIP-style teacher models, ranging from standard baselines to large-scale state-of-the-art models. Contrary to trends observed in NLP and vision, we find that stronger teachers do not consistently yield better students; in fact, existing distillation frameworks often fail to scale, leading to degraded performance in downstream multimodal tasks such as visual question answering. Our findings challenge prevailing assumptions in KD and point toward new directions for designing parameter-efficient multimodal models.
title When Better Teachers Don't Make Better Students: Revisiting Knowledge Distillation for CLIP Models in VQA
topic Computer Vision and Pattern Recognition
Computation and Language
url https://arxiv.org/abs/2511.17886