{"ID":2862783,"CreatedAt":"2026-06-01T04:54:23.091178241Z","UpdatedAt":"2026-06-01T04:54:23.091178241Z","DeletedAt":null,"paper_url":"https://arxiv.org/abs/2509.26235","arxiv_id":"2509.26235","title":"Interpret, prune and distill Donut : towards lightweight VLMs for VQA on document","abstract":"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.","short_abstract":"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 investiga...","url_abs":"https://arxiv.org/abs/2509.26235","url_pdf":"https://arxiv.org/pdf/2509.26235v1","authors":"[\"Adnan Ben Mansour\",\"Ayoub Karine\",\"David Naccache\"]","published":"2025-09-30T13:31:03Z","proceeding":"cs.CV","tasks":"[\"cs.CV\"]","methods":"[\"Language Model\"]","has_code":false}