{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:53QNGHKK4LAJJZTNF7BRLFJSY6","short_pith_number":"pith:53QNGHKK","schema_version":"1.0","canonical_sha256":"eee0d31d4ae2c094e66d2fc3159532c7862fa52a585c47be5788844cc7dc5649","source":{"kind":"arxiv","id":"2507.20749","version":1},"attestation_state":"computed","paper":{"title":"Investigating Structural Pruning and Recovery Techniques for Compressing Multimodal Large Language Models: An Empirical Study","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.CL","authors_text":"Lukas Thede, Massimiliano Mancini, Wenjia Xu, Yiran Huang, Zeynep Akata","submitted_at":"2025-07-28T11:57:52Z","abstract_excerpt":"While Multimodal Large Language Models (MLLMs) demonstrate impressive capabilities, their substantial computational and memory requirements pose significant barriers to practical deployment. Current parameter reduction techniques primarily involve training MLLMs from Small Language Models (SLMs), but these methods offer limited flexibility and remain computationally intensive. To address this gap, we propose to directly compress existing MLLMs through structural pruning combined with efficient recovery training. Specifically, we investigate two structural pruning paradigms--layerwise and width"},"verification_status":{"content_addressed":true,"pith_receipt":true,"author_attested":false,"weak_author_claims":0,"strong_author_claims":0,"externally_anchored":false,"storage_verified":false,"citation_signatures":0,"replication_records":0,"graph_snapshot":true,"references_resolved":false,"formal_links_present":false},"canonical_record":{"source":{"id":"2507.20749","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.CL","submitted_at":"2025-07-28T11:57:52Z","cross_cats_sorted":["cs.CV"],"title_canon_sha256":"9c04b2f213f95fc11d59b6818f65a0383f595d5498cdd5d5d55d5160e890f4a8","abstract_canon_sha256":"b3dfef994b9ba285d6620fd6232e06561935915408ee4d2aa67e6fd400c29f39"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:44:25.240680Z","signature_b64":"cnBPwUi0CrDFcD6BCGA/wAwyHbvkfR/59cwZeoJL+kwz28L3pOq4BBMFkJ2y8UmegyNsvkpRdKjilKPWx2LuCw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"eee0d31d4ae2c094e66d2fc3159532c7862fa52a585c47be5788844cc7dc5649","last_reissued_at":"2026-07-05T11:44:25.240306Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:44:25.240306Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Investigating Structural Pruning and Recovery Techniques for Compressing Multimodal Large Language Models: An Empirical Study","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.CV"],"primary_cat":"cs.CL","authors_text":"Lukas Thede, Massimiliano Mancini, Wenjia Xu, Yiran Huang, Zeynep Akata","submitted_at":"2025-07-28T11:57:52Z","abstract_excerpt":"While Multimodal Large Language Models (MLLMs) demonstrate impressive capabilities, their substantial computational and memory requirements pose significant barriers to practical deployment. Current parameter reduction techniques primarily involve training MLLMs from Small Language Models (SLMs), but these methods offer limited flexibility and remain computationally intensive. To address this gap, we propose to directly compress existing MLLMs through structural pruning combined with efficient recovery training. Specifically, we investigate two structural pruning paradigms--layerwise and width"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2507.20749","kind":"arxiv","version":1},"verdict":{"id":null,"model_set":{},"created_at":null,"strongest_claim":"","one_line_summary":"","pipeline_version":null,"weakest_assumption":"","pith_extraction_headline":""},"integrity":{"clean":true,"summary":{"advisory":0,"critical":0,"by_detector":{},"informational":0},"endpoint":"/pith/2507.20749/integrity.json","findings":[],"available":true,"detectors_run":[],"snapshot_sha256":"c28c3603d3b5d939e8dc4c7e95fa8dfce3d595e45f758748cecf8e644a296938"},"references":{"count":0,"sample":[],"resolved_work":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57","internal_anchors":0},"formal_canon":{"evidence_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"author_claims":{"count":0,"strong_count":0,"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"builder_version":"pith-number-builder-2026-05-17-v1"},"aliases":[{"alias_kind":"arxiv","alias_value":"2507.20749","created_at":"2026-07-05T11:44:25.240374+00:00"},{"alias_kind":"arxiv_version","alias_value":"2507.20749v1","created_at":"2026-07-05T11:44:25.240374+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2507.20749","created_at":"2026-07-05T11:44:25.240374+00:00"},{"alias_kind":"pith_short_12","alias_value":"53QNGHKK4LAJ","created_at":"2026-07-05T11:44:25.240374+00:00"},{"alias_kind":"pith_short_16","alias_value":"53QNGHKK4LAJJZTN","created_at":"2026-07-05T11:44:25.240374+00:00"},{"alias_kind":"pith_short_8","alias_value":"53QNGHKK","created_at":"2026-07-05T11:44:25.240374+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/53QNGHKK4LAJJZTNF7BRLFJSY6","json":"https://pith.science/pith/53QNGHKK4LAJJZTNF7BRLFJSY6.json","graph_json":"https://pith.science/api/pith-number/53QNGHKK4LAJJZTNF7BRLFJSY6/graph.json","events_json":"https://pith.science/api/pith-number/53QNGHKK4LAJJZTNF7BRLFJSY6/events.json","paper":"https://pith.science/paper/53QNGHKK"},"agent_actions":{"view_html":"https://pith.science/pith/53QNGHKK4LAJJZTNF7BRLFJSY6","download_json":"https://pith.science/pith/53QNGHKK4LAJJZTNF7BRLFJSY6.json","view_paper":"https://pith.science/paper/53QNGHKK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2507.20749&json=true","fetch_graph":"https://pith.science/api/pith-number/53QNGHKK4LAJJZTNF7BRLFJSY6/graph.json","fetch_events":"https://pith.science/api/pith-number/53QNGHKK4LAJJZTNF7BRLFJSY6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/53QNGHKK4LAJJZTNF7BRLFJSY6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/53QNGHKK4LAJJZTNF7BRLFJSY6/action/storage_attestation","attest_author":"https://pith.science/pith/53QNGHKK4LAJJZTNF7BRLFJSY6/action/author_attestation","sign_citation":"https://pith.science/pith/53QNGHKK4LAJJZTNF7BRLFJSY6/action/citation_signature","submit_replication":"https://pith.science/pith/53QNGHKK4LAJJZTNF7BRLFJSY6/action/replication_record"}},"created_at":"2026-07-05T11:44:25.240374+00:00","updated_at":"2026-07-05T11:44:25.240374+00:00"}