{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:QVMLCX7CWO3OI2FU57SUFYJI7F","short_pith_number":"pith:QVMLCX7C","schema_version":"1.0","canonical_sha256":"8558b15fe2b3b6e468b4efe542e128f940d20a0779c937ae6d721750a65ae388","source":{"kind":"arxiv","id":"2606.14150","version":2},"attestation_state":"computed","paper":{"title":"Small LLMs: Pruning vs. Training from Scratch","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Jiachen Zhu, Kunjun Li, Mingjie Sun, Taiming Lu, Yufeng Xu, Zhuang Liu","submitted_at":"2026-06-12T06:24:28Z","abstract_excerpt":"Pruning promises a shortcut to strong small language models. In this work, we examine this promise by pruning Llama-3.1-8B at pruning ratios of 0.5--0.8 with six methods spanning depth, width, and sparse granularities, under two controlled token-matched settings. (1) With the same training token budget, pruned initialization consistently outperforms random initialization. This shows that the parent model provides a strong starting point, although the advantage narrows as the training token budget grows and as the pruning ratio rises, nearly vanishing at the highest pruning ratio we study. (2) "},"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":"2606.14150","kind":"arxiv","version":2},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2026-06-12T06:24:28Z","cross_cats_sorted":["cs.CL"],"title_canon_sha256":"394fa4ce87186d75e6383d97a2f38eb414a1c2da65c898cf4ad34834ea4b95df","abstract_canon_sha256":"e79a911d0dddc9f27d349624a02b3ee967e4feac35abc0dcdda24b67a07c8425"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-06-23T01:13:06.700798Z","signature_b64":"kpiRWFvyQxPuCjL8bvcHhebTJWS0YHUesownOBkqpTeF2J6zX2eHtObP1SIwZmAxU1BCQ1T93DEqjNY0eezICA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"8558b15fe2b3b6e468b4efe542e128f940d20a0779c937ae6d721750a65ae388","last_reissued_at":"2026-06-23T01:13:06.700299Z","signature_status":"signed_v1","first_computed_at":"2026-06-23T01:13:06.700299Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Small LLMs: Pruning vs. Training from Scratch","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":["cs.CL"],"primary_cat":"cs.LG","authors_text":"Jiachen Zhu, Kunjun Li, Mingjie Sun, Taiming Lu, Yufeng Xu, Zhuang Liu","submitted_at":"2026-06-12T06:24:28Z","abstract_excerpt":"Pruning promises a shortcut to strong small language models. In this work, we examine this promise by pruning Llama-3.1-8B at pruning ratios of 0.5--0.8 with six methods spanning depth, width, and sparse granularities, under two controlled token-matched settings. (1) With the same training token budget, pruned initialization consistently outperforms random initialization. This shows that the parent model provides a strong starting point, although the advantage narrows as the training token budget grows and as the pruning ratio rises, nearly vanishing at the highest pruning ratio we study. (2) "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2606.14150","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/2606.14150/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":"2606.14150","created_at":"2026-06-23T01:13:06.700388+00:00"},{"alias_kind":"arxiv_version","alias_value":"2606.14150v2","created_at":"2026-06-23T01:13:06.700388+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2606.14150","created_at":"2026-06-23T01:13:06.700388+00:00"},{"alias_kind":"pith_short_12","alias_value":"QVMLCX7CWO3O","created_at":"2026-06-23T01:13:06.700388+00:00"},{"alias_kind":"pith_short_16","alias_value":"QVMLCX7CWO3OI2FU","created_at":"2026-06-23T01:13:06.700388+00:00"},{"alias_kind":"pith_short_8","alias_value":"QVMLCX7C","created_at":"2026-06-23T01:13:06.700388+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/QVMLCX7CWO3OI2FU57SUFYJI7F","json":"https://pith.science/pith/QVMLCX7CWO3OI2FU57SUFYJI7F.json","graph_json":"https://pith.science/api/pith-number/QVMLCX7CWO3OI2FU57SUFYJI7F/graph.json","events_json":"https://pith.science/api/pith-number/QVMLCX7CWO3OI2FU57SUFYJI7F/events.json","paper":"https://pith.science/paper/QVMLCX7C"},"agent_actions":{"view_html":"https://pith.science/pith/QVMLCX7CWO3OI2FU57SUFYJI7F","download_json":"https://pith.science/pith/QVMLCX7CWO3OI2FU57SUFYJI7F.json","view_paper":"https://pith.science/paper/QVMLCX7C","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2606.14150&json=true","fetch_graph":"https://pith.science/api/pith-number/QVMLCX7CWO3OI2FU57SUFYJI7F/graph.json","fetch_events":"https://pith.science/api/pith-number/QVMLCX7CWO3OI2FU57SUFYJI7F/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/QVMLCX7CWO3OI2FU57SUFYJI7F/action/timestamp_anchor","attest_storage":"https://pith.science/pith/QVMLCX7CWO3OI2FU57SUFYJI7F/action/storage_attestation","attest_author":"https://pith.science/pith/QVMLCX7CWO3OI2FU57SUFYJI7F/action/author_attestation","sign_citation":"https://pith.science/pith/QVMLCX7CWO3OI2FU57SUFYJI7F/action/citation_signature","submit_replication":"https://pith.science/pith/QVMLCX7CWO3OI2FU57SUFYJI7F/action/replication_record"}},"created_at":"2026-06-23T01:13:06.700388+00:00","updated_at":"2026-06-23T01:13:06.700388+00:00"}