{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:7MKHZY73OV5V47EZGVMDV6KFUV","short_pith_number":"pith:7MKHZY73","schema_version":"1.0","canonical_sha256":"fb147ce3fb757b5e7c9935583af945a55c5b54a2d9f7dcb0057b7f176dddd65d","source":{"kind":"arxiv","id":"2603.15389","version":2},"attestation_state":"computed","paper":{"title":"When Does Sparsity Mitigate the Curse of Depth in LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dilxat Muhtar, Max Zimmer, Nico Pelleriti, Sebastian Pokutta, Shiwei Liu, Thomas Hofmann, Xinyuan Song","submitted_at":"2026-03-16T15:04:16Z","abstract_excerpt":"Recent work has demonstrated the curse of depth in large language models (LLMs), where later layers contribute less to learning and representation than earlier layers. Such under-utilization is linked to the accumulated growth of variance in Pre-Layer Normalization, which can push deep blocks toward near-identity behavior. In this paper, we provide evidence that sparsity-like mechanisms can dampen variance propagation and are associated with improved depth utilization Our investigation covers two sources of sparsity: (i) implicit sparsity, which emerges from training and data conditions, inclu"},"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":"2603.15389","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.CL","submitted_at":"2026-03-16T15:04:16Z","cross_cats_sorted":[],"title_canon_sha256":"3ff0ef652af33fcfc4e731235f9f8ab8ad1631307b1ad032fd6a071a64f9a480","abstract_canon_sha256":"dd582aeec9f5c2a97e2001286952dc4f09532d296c3646658fa0fabdbe205904"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-30T01:16:29.278172Z","signature_b64":"xwoBUawD5MRMVaN6ObPJtE1yd7ZGXgbJ+LZm23n+s7UV4PNRkXxf4M00T/7sYCIkf7adAIu+3SHzSeGCpa2nAA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"fb147ce3fb757b5e7c9935583af945a55c5b54a2d9f7dcb0057b7f176dddd65d","last_reissued_at":"2026-06-30T01:16:29.277646Z","signature_status":"signed_v1","first_computed_at":"2026-06-30T01:16:29.277646Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"When Does Sparsity Mitigate the Curse of Depth in LLMs","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Dilxat Muhtar, Max Zimmer, Nico Pelleriti, Sebastian Pokutta, Shiwei Liu, Thomas Hofmann, Xinyuan Song","submitted_at":"2026-03-16T15:04:16Z","abstract_excerpt":"Recent work has demonstrated the curse of depth in large language models (LLMs), where later layers contribute less to learning and representation than earlier layers. Such under-utilization is linked to the accumulated growth of variance in Pre-Layer Normalization, which can push deep blocks toward near-identity behavior. In this paper, we provide evidence that sparsity-like mechanisms can dampen variance propagation and are associated with improved depth utilization Our investigation covers two sources of sparsity: (i) implicit sparsity, which emerges from training and data conditions, inclu"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2603.15389","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/2603.15389/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":"2603.15389","created_at":"2026-06-30T01:16:29.277717+00:00"},{"alias_kind":"arxiv_version","alias_value":"2603.15389v2","created_at":"2026-06-30T01:16:29.277717+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2603.15389","created_at":"2026-06-30T01:16:29.277717+00:00"},{"alias_kind":"pith_short_12","alias_value":"7MKHZY73OV5V","created_at":"2026-06-30T01:16:29.277717+00:00"},{"alias_kind":"pith_short_16","alias_value":"7MKHZY73OV5V47EZ","created_at":"2026-06-30T01:16:29.277717+00:00"},{"alias_kind":"pith_short_8","alias_value":"7MKHZY73","created_at":"2026-06-30T01:16:29.277717+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":3,"sample":[{"citing_arxiv_id":"2602.21750","citing_title":"From Words to Amino Acids: Does the Curse of Depth Persist?","ref_index":25,"is_internal_anchor":true},{"citing_arxiv_id":"2605.01725","citing_title":"Motion-Aware Caching for Efficient Autoregressive Video Generation","ref_index":44,"is_internal_anchor":true},{"citing_arxiv_id":"2605.12492","citing_title":"Pion: A Spectrum-Preserving Optimizer via Orthogonal Equivalence Transformation","ref_index":56,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/7MKHZY73OV5V47EZGVMDV6KFUV","json":"https://pith.science/pith/7MKHZY73OV5V47EZGVMDV6KFUV.json","graph_json":"https://pith.science/api/pith-number/7MKHZY73OV5V47EZGVMDV6KFUV/graph.json","events_json":"https://pith.science/api/pith-number/7MKHZY73OV5V47EZGVMDV6KFUV/events.json","paper":"https://pith.science/paper/7MKHZY73"},"agent_actions":{"view_html":"https://pith.science/pith/7MKHZY73OV5V47EZGVMDV6KFUV","download_json":"https://pith.science/pith/7MKHZY73OV5V47EZGVMDV6KFUV.json","view_paper":"https://pith.science/paper/7MKHZY73","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2603.15389&json=true","fetch_graph":"https://pith.science/api/pith-number/7MKHZY73OV5V47EZGVMDV6KFUV/graph.json","fetch_events":"https://pith.science/api/pith-number/7MKHZY73OV5V47EZGVMDV6KFUV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/7MKHZY73OV5V47EZGVMDV6KFUV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/7MKHZY73OV5V47EZGVMDV6KFUV/action/storage_attestation","attest_author":"https://pith.science/pith/7MKHZY73OV5V47EZGVMDV6KFUV/action/author_attestation","sign_citation":"https://pith.science/pith/7MKHZY73OV5V47EZGVMDV6KFUV/action/citation_signature","submit_replication":"https://pith.science/pith/7MKHZY73OV5V47EZGVMDV6KFUV/action/replication_record"}},"created_at":"2026-06-30T01:16:29.277717+00:00","updated_at":"2026-06-30T01:16:29.277717+00:00"}