{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2021:IE5AYRLKXPWXQAKDIA7VJZFBG6","short_pith_number":"pith:IE5AYRLK","schema_version":"1.0","canonical_sha256":"413a0c456abbed780143403f54e4a1378e2522d624a09193f8ace323fa792159","source":{"kind":"arxiv","id":"2102.01386","version":2},"attestation_state":"computed","paper":{"title":"AutoFreeze: Automatically Freezing Model Blocks to Accelerate Fine-tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Saurabh Agarwal, Shivaram Venkataraman, Yuhan Liu","submitted_at":"2021-02-02T08:35:02Z","abstract_excerpt":"With the rapid adoption of machine learning (ML), a number of domains now use the approach of fine tuning models which were pre-trained on a large corpus of data. However, our experiments show that even fine-tuning on models like BERT can take many hours even when using modern accelerators like GPUs. While prior work proposes limiting the number of layers that are fine-tuned, e.g., freezing all layers but the last layer, we find that such static approaches lead to reduced accuracy. We propose, AutoFreeze, a system that uses an adaptive approach to choose which layers are trained and show how 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":"2102.01386","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.LG","submitted_at":"2021-02-02T08:35:02Z","cross_cats_sorted":[],"title_canon_sha256":"3c508e316ca55bbcfca33aa0a183f7e41673c16ccb37f5f748f2219bde0e26e0","abstract_canon_sha256":"e564a364b7e0fe99ed564846a62cc918cf01bf3be560237a4a1bbd759e1d6889"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T02:28:55.467089Z","signature_b64":"idqrmvPREKIpXgJH/77KPAvs7PmTlw6mzT6CkGxEpCkr2k1Ka8WBQbj0AocZZPuXABcmxT/zMv/BVaGlKrw8Aw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"413a0c456abbed780143403f54e4a1378e2522d624a09193f8ace323fa792159","last_reissued_at":"2026-07-05T02:28:55.466579Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T02:28:55.466579Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"AutoFreeze: Automatically Freezing Model Blocks to Accelerate Fine-tuning","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Saurabh Agarwal, Shivaram Venkataraman, Yuhan Liu","submitted_at":"2021-02-02T08:35:02Z","abstract_excerpt":"With the rapid adoption of machine learning (ML), a number of domains now use the approach of fine tuning models which were pre-trained on a large corpus of data. However, our experiments show that even fine-tuning on models like BERT can take many hours even when using modern accelerators like GPUs. While prior work proposes limiting the number of layers that are fine-tuned, e.g., freezing all layers but the last layer, we find that such static approaches lead to reduced accuracy. We propose, AutoFreeze, a system that uses an adaptive approach to choose which layers are trained and show how t"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2102.01386","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/2102.01386/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":"2102.01386","created_at":"2026-07-05T02:28:55.466640+00:00"},{"alias_kind":"arxiv_version","alias_value":"2102.01386v2","created_at":"2026-07-05T02:28:55.466640+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2102.01386","created_at":"2026-07-05T02:28:55.466640+00:00"},{"alias_kind":"pith_short_12","alias_value":"IE5AYRLKXPWX","created_at":"2026-07-05T02:28:55.466640+00:00"},{"alias_kind":"pith_short_16","alias_value":"IE5AYRLKXPWXQAKD","created_at":"2026-07-05T02:28:55.466640+00:00"},{"alias_kind":"pith_short_8","alias_value":"IE5AYRLK","created_at":"2026-07-05T02:28:55.466640+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":2,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.17526","citing_title":"MGUP: A Momentum-Gradient Alignment Update Policy for Stochastic Optimization","ref_index":6,"is_internal_anchor":false},{"citing_arxiv_id":"2512.04844","citing_title":"Mitigating Catastrophic Forgetting in Target Language Adaptation of LLMs via Source-Shielded Updates","ref_index":51,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/IE5AYRLKXPWXQAKDIA7VJZFBG6","json":"https://pith.science/pith/IE5AYRLKXPWXQAKDIA7VJZFBG6.json","graph_json":"https://pith.science/api/pith-number/IE5AYRLKXPWXQAKDIA7VJZFBG6/graph.json","events_json":"https://pith.science/api/pith-number/IE5AYRLKXPWXQAKDIA7VJZFBG6/events.json","paper":"https://pith.science/paper/IE5AYRLK"},"agent_actions":{"view_html":"https://pith.science/pith/IE5AYRLKXPWXQAKDIA7VJZFBG6","download_json":"https://pith.science/pith/IE5AYRLKXPWXQAKDIA7VJZFBG6.json","view_paper":"https://pith.science/paper/IE5AYRLK","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2102.01386&json=true","fetch_graph":"https://pith.science/api/pith-number/IE5AYRLKXPWXQAKDIA7VJZFBG6/graph.json","fetch_events":"https://pith.science/api/pith-number/IE5AYRLKXPWXQAKDIA7VJZFBG6/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/IE5AYRLKXPWXQAKDIA7VJZFBG6/action/timestamp_anchor","attest_storage":"https://pith.science/pith/IE5AYRLKXPWXQAKDIA7VJZFBG6/action/storage_attestation","attest_author":"https://pith.science/pith/IE5AYRLKXPWXQAKDIA7VJZFBG6/action/author_attestation","sign_citation":"https://pith.science/pith/IE5AYRLKXPWXQAKDIA7VJZFBG6/action/citation_signature","submit_replication":"https://pith.science/pith/IE5AYRLKXPWXQAKDIA7VJZFBG6/action/replication_record"}},"created_at":"2026-07-05T02:28:55.466640+00:00","updated_at":"2026-07-05T02:28:55.466640+00:00"}