Interpret, prune and distill Donut : towards lightweight VLMs for VQA on document

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Mansour, Adnan Ben, Karine, Ayoub, Naccache, David
Format: Preprint
Veröffentlicht: 2025
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866915525176590336
author Mansour, Adnan Ben
Karine, Ayoub
Naccache, David
author_facet Mansour, Adnan Ben
Karine, Ayoub
Naccache, David
contents Recent advances in Visually-rich Document Understanding rely on large Vision-Language Models like Donut, which perform document-level Visual Question Answering without Optical Character Recognition. Despite their effectiveness, these models are too costly for real-time or resource-constrained applications. We investigate model compression through knowledge distillation, training compact student models from a larger teacher. We leverage mechanistic interpretability to drive student architecture design within this framework. By analyzing internal computations, we identify essential subcomponents to retain, while having a clear view of which subcomponents should be approximated, skipped, or reparametrized based on their function. This approach yields Donut-MINT (Mechanistic Interpretability-based Network Trimming), a pruned Donut variant that reduces inference time and memory usage while maintaining strong performance on DocVQA, a standard benchmark for document Visual Question Answering. Our method reframes compression as circuit discovery, bridging interpretability research and practical Vision-Language Model deployment.
format Preprint
id arxiv_https___arxiv_org_abs_2509_26235
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Interpret, prune and distill Donut : towards lightweight VLMs for VQA on document
Mansour, Adnan Ben
Karine, Ayoub
Naccache, David
Computer Vision and Pattern Recognition
Recent advances in Visually-rich Document Understanding rely on large Vision-Language Models like Donut, which perform document-level Visual Question Answering without Optical Character Recognition. Despite their effectiveness, these models are too costly for real-time or resource-constrained applications. We investigate model compression through knowledge distillation, training compact student models from a larger teacher. We leverage mechanistic interpretability to drive student architecture design within this framework. By analyzing internal computations, we identify essential subcomponents to retain, while having a clear view of which subcomponents should be approximated, skipped, or reparametrized based on their function. This approach yields Donut-MINT (Mechanistic Interpretability-based Network Trimming), a pruned Donut variant that reduces inference time and memory usage while maintaining strong performance on DocVQA, a standard benchmark for document Visual Question Answering. Our method reframes compression as circuit discovery, bridging interpretability research and practical Vision-Language Model deployment.
title Interpret, prune and distill Donut : towards lightweight VLMs for VQA on document
topic Computer Vision and Pattern Recognition
url https://arxiv.org/abs/2509.26235