{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:FQTITAF5REVN3EX6H2BULKNA5F","short_pith_number":"pith:FQTITAF5","schema_version":"1.0","canonical_sha256":"2c268980bd892add92fe3e8345a9a0e9572e93f58bf59eab3643df2e531f2beb","source":{"kind":"arxiv","id":"2509.23729","version":3},"attestation_state":"computed","paper":{"title":"LUQ: Layerwise Ultra-Low Bit Quantization for Multimodal Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG","eess.IV"],"primary_cat":"cs.CV","authors_text":"Andy Xu, Kar-Han Tan, Narendra Ahuja, Shubhang Bhatnagar","submitted_at":"2025-09-28T08:20:00Z","abstract_excerpt":"Large Language Models (LLMs) with multimodal capabilities have revolutionized vision-language tasks, but their deployment often requires huge memory and computational resources. Post-training quantization (PTQ) has successfully compressed language models to as low as 1-bit precision, its effectiveness for multimodal LLMs (MLLMs) remains unexplored. In this paper, we present the first method for ultra-low-bit (<4-bit) quantization of MLLMs. Our analysis reveals that multimodal tokens and intermediate layer activations produced by them exhibit significantly higher entropy compared to text tokens"},"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":"2509.23729","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CV","submitted_at":"2025-09-28T08:20:00Z","cross_cats_sorted":["cs.AI","cs.LG","eess.IV"],"title_canon_sha256":"424d82d822361fa329fd3dbbef88a9464bdb8ef595866f345d3e45ff8bb25bd6","abstract_canon_sha256":"4488913876fdeef2225c9aa55d9b735ef5f6493b980376f89421964ed213fb7f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-23T02:13:17.120295Z","signature_b64":"1Ol4J27QN+e2Aoi1qTDIcw8772/p2CfQRa8JSoImbNjrD+7Nt6p+NfYuuskux5OHAlOjiNS1BnseMLFvJUDQBQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"2c268980bd892add92fe3e8345a9a0e9572e93f58bf59eab3643df2e531f2beb","last_reissued_at":"2026-06-23T02:13:17.119780Z","signature_status":"signed_v1","first_computed_at":"2026-06-23T02:13:17.119780Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"LUQ: Layerwise Ultra-Low Bit Quantization for Multimodal Large Language Models","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.AI","cs.LG","eess.IV"],"primary_cat":"cs.CV","authors_text":"Andy Xu, Kar-Han Tan, Narendra Ahuja, Shubhang Bhatnagar","submitted_at":"2025-09-28T08:20:00Z","abstract_excerpt":"Large Language Models (LLMs) with multimodal capabilities have revolutionized vision-language tasks, but their deployment often requires huge memory and computational resources. Post-training quantization (PTQ) has successfully compressed language models to as low as 1-bit precision, its effectiveness for multimodal LLMs (MLLMs) remains unexplored. In this paper, we present the first method for ultra-low-bit (<4-bit) quantization of MLLMs. Our analysis reveals that multimodal tokens and intermediate layer activations produced by them exhibit significantly higher entropy compared to text tokens"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2509.23729","kind":"arxiv","version":3},"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/2509.23729/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":"2509.23729","created_at":"2026-06-23T02:13:17.119846+00:00"},{"alias_kind":"arxiv_version","alias_value":"2509.23729v3","created_at":"2026-06-23T02:13:17.119846+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2509.23729","created_at":"2026-06-23T02:13:17.119846+00:00"},{"alias_kind":"pith_short_12","alias_value":"FQTITAF5REVN","created_at":"2026-06-23T02:13:17.119846+00:00"},{"alias_kind":"pith_short_16","alias_value":"FQTITAF5REVN3EX6","created_at":"2026-06-23T02:13:17.119846+00:00"},{"alias_kind":"pith_short_8","alias_value":"FQTITAF5","created_at":"2026-06-23T02:13:17.119846+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2604.17320","citing_title":"Towards Joint Quantization and Token Pruning of Vision-Language Models","ref_index":3,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/FQTITAF5REVN3EX6H2BULKNA5F","json":"https://pith.science/pith/FQTITAF5REVN3EX6H2BULKNA5F.json","graph_json":"https://pith.science/api/pith-number/FQTITAF5REVN3EX6H2BULKNA5F/graph.json","events_json":"https://pith.science/api/pith-number/FQTITAF5REVN3EX6H2BULKNA5F/events.json","paper":"https://pith.science/paper/FQTITAF5"},"agent_actions":{"view_html":"https://pith.science/pith/FQTITAF5REVN3EX6H2BULKNA5F","download_json":"https://pith.science/pith/FQTITAF5REVN3EX6H2BULKNA5F.json","view_paper":"https://pith.science/paper/FQTITAF5","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2509.23729&json=true","fetch_graph":"https://pith.science/api/pith-number/FQTITAF5REVN3EX6H2BULKNA5F/graph.json","fetch_events":"https://pith.science/api/pith-number/FQTITAF5REVN3EX6H2BULKNA5F/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/FQTITAF5REVN3EX6H2BULKNA5F/action/timestamp_anchor","attest_storage":"https://pith.science/pith/FQTITAF5REVN3EX6H2BULKNA5F/action/storage_attestation","attest_author":"https://pith.science/pith/FQTITAF5REVN3EX6H2BULKNA5F/action/author_attestation","sign_citation":"https://pith.science/pith/FQTITAF5REVN3EX6H2BULKNA5F/action/citation_signature","submit_replication":"https://pith.science/pith/FQTITAF5REVN3EX6H2BULKNA5F/action/replication_record"}},"created_at":"2026-06-23T02:13:17.119846+00:00","updated_at":"2026-06-23T02:13:17.119846+00:00"}