{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:36ATZ4OLCWHN6GFQCUT5AY3H24","short_pith_number":"pith:36ATZ4OL","schema_version":"1.0","canonical_sha256":"df813cf1cb158edf18b01527d06367d7337cd257a2d59bd723801703880b1043","source":{"kind":"arxiv","id":"2501.15674","version":2},"attestation_state":"computed","paper":{"title":"TensorLLM: Tensorising Multi-Head Attention for Enhanced Reasoning and Compression in LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Danilo Mandic, Giorgos Iacovides, Wuyang Zhou, Yuxuan Gu","submitted_at":"2025-01-26T21:05:16Z","abstract_excerpt":"The reasoning abilities of Large Language Models (LLMs) can be improved by structurally denoising their weights, yet existing techniques primarily focus on denoising the feed-forward network (FFN) of the transformer block, and can not efficiently utilise the Multi-head Attention (MHA) block, which is the core of transformer architectures. To address this issue, we propose a novel intuitive framework that, at its very core, performs MHA compression through a multi-head tensorisation process and the Tucker decomposition. This enables both higher-dimensional structured denoising and compression o"},"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":"2501.15674","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2025-01-26T21:05:16Z","cross_cats_sorted":["cs.LG"],"title_canon_sha256":"1febe1f08b69979c7599ab2ebb2b5d5b8a41c31ddf819caa8ba1af697ca2cdd0","abstract_canon_sha256":"58ca45f319b64a328ec77635fa3c8de20821eaf31e30bebe075c4ad910c8b1ea"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:03:20.786055Z","signature_b64":"wDgr6eKHbsGP7W+WmzcJ1Zrxo/0PPMHpZ1l7qupfBiRs9cECASfq54hYeNDAHrnTqR2fDqJ6fkev4Kc08pGyCg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"df813cf1cb158edf18b01527d06367d7337cd257a2d59bd723801703880b1043","last_reissued_at":"2026-07-05T11:03:20.785570Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:03:20.785570Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"TensorLLM: Tensorising Multi-Head Attention for Enhanced Reasoning and Compression in LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.LG"],"primary_cat":"cs.CL","authors_text":"Danilo Mandic, Giorgos Iacovides, Wuyang Zhou, Yuxuan Gu","submitted_at":"2025-01-26T21:05:16Z","abstract_excerpt":"The reasoning abilities of Large Language Models (LLMs) can be improved by structurally denoising their weights, yet existing techniques primarily focus on denoising the feed-forward network (FFN) of the transformer block, and can not efficiently utilise the Multi-head Attention (MHA) block, which is the core of transformer architectures. To address this issue, we propose a novel intuitive framework that, at its very core, performs MHA compression through a multi-head tensorisation process and the Tucker decomposition. This enables both higher-dimensional structured denoising and compression o"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2501.15674","kind":"arxiv","version":2},"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/2501.15674/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":"2501.15674","created_at":"2026-07-05T11:03:20.785627+00:00"},{"alias_kind":"arxiv_version","alias_value":"2501.15674v2","created_at":"2026-07-05T11:03:20.785627+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2501.15674","created_at":"2026-07-05T11:03:20.785627+00:00"},{"alias_kind":"pith_short_12","alias_value":"36ATZ4OLCWHN","created_at":"2026-07-05T11:03:20.785627+00:00"},{"alias_kind":"pith_short_16","alias_value":"36ATZ4OLCWHN6GFQ","created_at":"2026-07-05T11:03:20.785627+00:00"},{"alias_kind":"pith_short_8","alias_value":"36ATZ4OL","created_at":"2026-07-05T11:03:20.785627+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.03465","citing_title":"Rethinking the Role of Tensor Decompositions in Post-Training LLM Compression","ref_index":16,"is_internal_anchor":false},{"citing_arxiv_id":"2509.03234","citing_title":"TeRA: Vector-based Random Tensor Network for High-Rank Adaptation of Large Language Models","ref_index":16,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/36ATZ4OLCWHN6GFQCUT5AY3H24","json":"https://pith.science/pith/36ATZ4OLCWHN6GFQCUT5AY3H24.json","graph_json":"https://pith.science/api/pith-number/36ATZ4OLCWHN6GFQCUT5AY3H24/graph.json","events_json":"https://pith.science/api/pith-number/36ATZ4OLCWHN6GFQCUT5AY3H24/events.json","paper":"https://pith.science/paper/36ATZ4OL"},"agent_actions":{"view_html":"https://pith.science/pith/36ATZ4OLCWHN6GFQCUT5AY3H24","download_json":"https://pith.science/pith/36ATZ4OLCWHN6GFQCUT5AY3H24.json","view_paper":"https://pith.science/paper/36ATZ4OL","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2501.15674&json=true","fetch_graph":"https://pith.science/api/pith-number/36ATZ4OLCWHN6GFQCUT5AY3H24/graph.json","fetch_events":"https://pith.science/api/pith-number/36ATZ4OLCWHN6GFQCUT5AY3H24/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/36ATZ4OLCWHN6GFQCUT5AY3H24/action/timestamp_anchor","attest_storage":"https://pith.science/pith/36ATZ4OLCWHN6GFQCUT5AY3H24/action/storage_attestation","attest_author":"https://pith.science/pith/36ATZ4OLCWHN6GFQCUT5AY3H24/action/author_attestation","sign_citation":"https://pith.science/pith/36ATZ4OLCWHN6GFQCUT5AY3H24/action/citation_signature","submit_replication":"https://pith.science/pith/36ATZ4OLCWHN6GFQCUT5AY3H24/action/replication_record"}},"created_at":"2026-07-05T11:03:20.785627+00:00","updated_at":"2026-07-05T11:03:20.785627+00:00"}