{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:M5UFH6VEXET3337XVLJX4ZIITZ","short_pith_number":"pith:M5UFH6VE","schema_version":"1.0","canonical_sha256":"676853faa4b927bdeff7aad37e65089e537af03d39857520f7773f9742c7ee5b","source":{"kind":"arxiv","id":"2505.11643","version":1},"attestation_state":"computed","paper":{"title":"Can an Easy-to-Hard Curriculum Make Reasoning Emerge in Small Language Models? Evidence from a Four-Stage Curriculum on GPT-2","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Xiang Fu","submitted_at":"2025-05-16T19:08:31Z","abstract_excerpt":"We demonstrate that a developmentally ordered curriculum markedly improves reasoning transparency and sample-efficiency in small language models (SLMs). Concretely, we train Cognivolve, a 124 M-parameter GPT-2 model, on a four-stage syllabus that ascends from lexical matching to multi-step symbolic inference and then evaluate it without any task-specific fine-tuning. Cognivolve reaches target accuracy in half the optimization steps of a single-phase baseline, activates an order-of-magnitude more gradient-salient reasoning heads, and shifts those heads toward deeper layers, yielding higher-entr"},"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":"2505.11643","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-05-16T19:08:31Z","cross_cats_sorted":[],"title_canon_sha256":"88f2af1ffe16c52859277aca19d2e065554aefe77ba5eba93da6dc43b7850f61","abstract_canon_sha256":"b436c0fc756183da842970f0f9fefc4971e024073feb0627eb2cb5619cf432ec"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:04:24.712001Z","signature_b64":"tbe/ab53Kx1oONC+yv7aFRfegUv/HPvUqT5/kuMqNMXw0yJW3QwqgOq8Ib1KXTBkrWgmGYiY8Ay9xkJP9O7dBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"676853faa4b927bdeff7aad37e65089e537af03d39857520f7773f9742c7ee5b","last_reissued_at":"2026-07-05T11:04:24.711464Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:04:24.711464Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Can an Easy-to-Hard Curriculum Make Reasoning Emerge in Small Language Models? Evidence from a Four-Stage Curriculum on GPT-2","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Xiang Fu","submitted_at":"2025-05-16T19:08:31Z","abstract_excerpt":"We demonstrate that a developmentally ordered curriculum markedly improves reasoning transparency and sample-efficiency in small language models (SLMs). Concretely, we train Cognivolve, a 124 M-parameter GPT-2 model, on a four-stage syllabus that ascends from lexical matching to multi-step symbolic inference and then evaluate it without any task-specific fine-tuning. Cognivolve reaches target accuracy in half the optimization steps of a single-phase baseline, activates an order-of-magnitude more gradient-salient reasoning heads, and shifts those heads toward deeper layers, yielding higher-entr"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.11643","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/2505.11643/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":"2505.11643","created_at":"2026-07-05T11:04:24.711527+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.11643v1","created_at":"2026-07-05T11:04:24.711527+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.11643","created_at":"2026-07-05T11:04:24.711527+00:00"},{"alias_kind":"pith_short_12","alias_value":"M5UFH6VEXET3","created_at":"2026-07-05T11:04:24.711527+00:00"},{"alias_kind":"pith_short_16","alias_value":"M5UFH6VEXET3337X","created_at":"2026-07-05T11:04:24.711527+00:00"},{"alias_kind":"pith_short_8","alias_value":"M5UFH6VE","created_at":"2026-07-05T11:04:24.711527+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2507.13966","citing_title":"Bottom-up Domain-specific Superintelligence: A Reliable Knowledge Graph is What We Need","ref_index":87,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/M5UFH6VEXET3337XVLJX4ZIITZ","json":"https://pith.science/pith/M5UFH6VEXET3337XVLJX4ZIITZ.json","graph_json":"https://pith.science/api/pith-number/M5UFH6VEXET3337XVLJX4ZIITZ/graph.json","events_json":"https://pith.science/api/pith-number/M5UFH6VEXET3337XVLJX4ZIITZ/events.json","paper":"https://pith.science/paper/M5UFH6VE"},"agent_actions":{"view_html":"https://pith.science/pith/M5UFH6VEXET3337XVLJX4ZIITZ","download_json":"https://pith.science/pith/M5UFH6VEXET3337XVLJX4ZIITZ.json","view_paper":"https://pith.science/paper/M5UFH6VE","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.11643&json=true","fetch_graph":"https://pith.science/api/pith-number/M5UFH6VEXET3337XVLJX4ZIITZ/graph.json","fetch_events":"https://pith.science/api/pith-number/M5UFH6VEXET3337XVLJX4ZIITZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/M5UFH6VEXET3337XVLJX4ZIITZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/M5UFH6VEXET3337XVLJX4ZIITZ/action/storage_attestation","attest_author":"https://pith.science/pith/M5UFH6VEXET3337XVLJX4ZIITZ/action/author_attestation","sign_citation":"https://pith.science/pith/M5UFH6VEXET3337XVLJX4ZIITZ/action/citation_signature","submit_replication":"https://pith.science/pith/M5UFH6VEXET3337XVLJX4ZIITZ/action/replication_record"}},"created_at":"2026-07-05T11:04:24.711527+00:00","updated_at":"2026-07-05T11:04:24.711527+00:00"}