{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:EMFBK6I2MVQ7DSL3RE6JNQ2WTJ","short_pith_number":"pith:EMFBK6I2","schema_version":"1.0","canonical_sha256":"230a15791a6561f1c97b893c96c3569a512902313b76ebb8dd6028b4d790e882","source":{"kind":"arxiv","id":"2505.15442","version":2},"attestation_state":"computed","paper":{"title":"On the Generalization vs Fidelity Paradox in Knowledge Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ayan Sengupta, Suhas Kamasetty Ramesh, Tanmoy Chakraborty","submitted_at":"2025-05-21T12:23:32Z","abstract_excerpt":"Knowledge distillation (KD) is a key technique for compressing large language models into smaller ones while preserving performance. Despite the recent traction of KD research, its effectiveness for smaller language models (LMs) and the mechanisms driving knowledge transfer remain underexplored. In this work, we present the first large-scale empirical and statistical analysis of KD across models ranging from 0.5B to 7B parameters on 14 complex reasoning tasks in a zero-shot setting. Our findings reveal that KD can improve the average performance of smaller models by up to $10\\%$, with a peak t"},"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":"2505.15442","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.CL","submitted_at":"2025-05-21T12:23:32Z","cross_cats_sorted":[],"title_canon_sha256":"bda27bc04382bf53d04ecd2173c1e93d77eaf96d45e83a460490e48287c1889e","abstract_canon_sha256":"526a9ce6e83b863a969786555421375625eed795b29ba369a902fdb921e68f9f"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T11:48:03.981695Z","signature_b64":"5Kr/O2yW3iHYsS7OCq54yDfIIYt+ffZvtoKHdRc0FpcmYVNv1zsZ5PGfhAht025i8PpyqDTksXsnjsfF4bPTDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"230a15791a6561f1c97b893c96c3569a512902313b76ebb8dd6028b4d790e882","last_reissued_at":"2026-07-05T11:48:03.981097Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T11:48:03.981097Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"On the Generalization vs Fidelity Paradox in Knowledge Distillation","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.CL","authors_text":"Ayan Sengupta, Suhas Kamasetty Ramesh, Tanmoy Chakraborty","submitted_at":"2025-05-21T12:23:32Z","abstract_excerpt":"Knowledge distillation (KD) is a key technique for compressing large language models into smaller ones while preserving performance. Despite the recent traction of KD research, its effectiveness for smaller language models (LMs) and the mechanisms driving knowledge transfer remain underexplored. In this work, we present the first large-scale empirical and statistical analysis of KD across models ranging from 0.5B to 7B parameters on 14 complex reasoning tasks in a zero-shot setting. Our findings reveal that KD can improve the average performance of smaller models by up to $10\\%$, with a peak t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2505.15442","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/2505.15442/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":"2505.15442","created_at":"2026-07-05T11:48:03.981155+00:00"},{"alias_kind":"arxiv_version","alias_value":"2505.15442v2","created_at":"2026-07-05T11:48:03.981155+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2505.15442","created_at":"2026-07-05T11:48:03.981155+00:00"},{"alias_kind":"pith_short_12","alias_value":"EMFBK6I2MVQ7","created_at":"2026-07-05T11:48:03.981155+00:00"},{"alias_kind":"pith_short_16","alias_value":"EMFBK6I2MVQ7DSL3","created_at":"2026-07-05T11:48:03.981155+00:00"},{"alias_kind":"pith_short_8","alias_value":"EMFBK6I2","created_at":"2026-07-05T11:48:03.981155+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":3,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2607.08268","citing_title":"Different Teachers, Different Capabilities: Sub-1B On-Device Distillation for Structured Text Enrichment","ref_index":14,"is_internal_anchor":true},{"citing_arxiv_id":"2606.06840","citing_title":"Characterize Then Distill: Mechanistic Reasoning in Large Output Spaces","ref_index":36,"is_internal_anchor":false},{"citing_arxiv_id":"2509.06337","citing_title":"Large Language Models as Virtual Survey Respondents: Evaluating Sociodemographic Response Generation","ref_index":28,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/EMFBK6I2MVQ7DSL3RE6JNQ2WTJ","json":"https://pith.science/pith/EMFBK6I2MVQ7DSL3RE6JNQ2WTJ.json","graph_json":"https://pith.science/api/pith-number/EMFBK6I2MVQ7DSL3RE6JNQ2WTJ/graph.json","events_json":"https://pith.science/api/pith-number/EMFBK6I2MVQ7DSL3RE6JNQ2WTJ/events.json","paper":"https://pith.science/paper/EMFBK6I2"},"agent_actions":{"view_html":"https://pith.science/pith/EMFBK6I2MVQ7DSL3RE6JNQ2WTJ","download_json":"https://pith.science/pith/EMFBK6I2MVQ7DSL3RE6JNQ2WTJ.json","view_paper":"https://pith.science/paper/EMFBK6I2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2505.15442&json=true","fetch_graph":"https://pith.science/api/pith-number/EMFBK6I2MVQ7DSL3RE6JNQ2WTJ/graph.json","fetch_events":"https://pith.science/api/pith-number/EMFBK6I2MVQ7DSL3RE6JNQ2WTJ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/EMFBK6I2MVQ7DSL3RE6JNQ2WTJ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/EMFBK6I2MVQ7DSL3RE6JNQ2WTJ/action/storage_attestation","attest_author":"https://pith.science/pith/EMFBK6I2MVQ7DSL3RE6JNQ2WTJ/action/author_attestation","sign_citation":"https://pith.science/pith/EMFBK6I2MVQ7DSL3RE6JNQ2WTJ/action/citation_signature","submit_replication":"https://pith.science/pith/EMFBK6I2MVQ7DSL3RE6JNQ2WTJ/action/replication_record"}},"created_at":"2026-07-05T11:48:03.981155+00:00","updated_at":"2026-07-05T11:48:03.981155+00:00"}