{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2023:UPHQIYP2QOJJVFM2EUGWH7M7RU","short_pith_number":"pith:UPHQIYP2","schema_version":"1.0","canonical_sha256":"a3cf0461fa83929a959a250d63fd9f8d095cbb771b3a34b89bbe87e9ebdbb5c9","source":{"kind":"arxiv","id":"2302.01172","version":1},"attestation_state":"computed","paper":{"title":"STEP: Learning N:M Structured Sparsity Masks from Scratch with Precondition","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Amir Yazdanbakhsh, Christopher De Sa, Oleg Rybakov, Shivani Agrawal, Suvinay Subramanian, Yucheng Lu","submitted_at":"2023-02-02T15:49:03Z","abstract_excerpt":"Recent innovations on hardware (e.g. Nvidia A100) have motivated learning N:M structured sparsity masks from scratch for fast model inference. However, state-of-the-art learning recipes in this regime (e.g. SR-STE) are proposed for non-adaptive optimizers like momentum SGD, while incurring non-trivial accuracy drop for Adam-trained models like attention-based LLMs. In this paper, we first demonstrate such gap origins from poorly estimated second moment (i.e. variance) in Adam states given by the masked weights. We conjecture that learning N:M masks with Adam should take the critical regime of "},"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":"2302.01172","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.LG","submitted_at":"2023-02-02T15:49:03Z","cross_cats_sorted":[],"title_canon_sha256":"229997eee134f0b9f5dca93b828e94f15008e3193b3682472df6857fc7fee3ea","abstract_canon_sha256":"38296cf27cd298e28f265c29e7825123daf2a08c2c57b091b6ab78546c98aacc"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T05:38:23.771628Z","signature_b64":"N3woKFpEzOf7VwMhCix9KfVqwBNZ0rfPKpwF9v1v3FOzqRCv4ujeAC3vYdGXlALkj9sZa1k8iX2GfAF5XJekCA==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a3cf0461fa83929a959a250d63fd9f8d095cbb771b3a34b89bbe87e9ebdbb5c9","last_reissued_at":"2026-07-05T05:38:23.771256Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T05:38:23.771256Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"STEP: Learning N:M Structured Sparsity Masks from Scratch with Precondition","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.LG","authors_text":"Amir Yazdanbakhsh, Christopher De Sa, Oleg Rybakov, Shivani Agrawal, Suvinay Subramanian, Yucheng Lu","submitted_at":"2023-02-02T15:49:03Z","abstract_excerpt":"Recent innovations on hardware (e.g. Nvidia A100) have motivated learning N:M structured sparsity masks from scratch for fast model inference. However, state-of-the-art learning recipes in this regime (e.g. SR-STE) are proposed for non-adaptive optimizers like momentum SGD, while incurring non-trivial accuracy drop for Adam-trained models like attention-based LLMs. In this paper, we first demonstrate such gap origins from poorly estimated second moment (i.e. variance) in Adam states given by the masked weights. We conjecture that learning N:M masks with Adam should take the critical regime of "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2302.01172","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/2302.01172/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":"2302.01172","created_at":"2026-07-05T05:38:23.771326+00:00"},{"alias_kind":"arxiv_version","alias_value":"2302.01172v1","created_at":"2026-07-05T05:38:23.771326+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2302.01172","created_at":"2026-07-05T05:38:23.771326+00:00"},{"alias_kind":"pith_short_12","alias_value":"UPHQIYP2QOJJ","created_at":"2026-07-05T05:38:23.771326+00:00"},{"alias_kind":"pith_short_16","alias_value":"UPHQIYP2QOJJVFM2","created_at":"2026-07-05T05:38:23.771326+00:00"},{"alias_kind":"pith_short_8","alias_value":"UPHQIYP2","created_at":"2026-07-05T05:38:23.771326+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2507.17301","citing_title":"Efficient Column-Wise N:M Pruning on RISC-V CPU","ref_index":28,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/UPHQIYP2QOJJVFM2EUGWH7M7RU","json":"https://pith.science/pith/UPHQIYP2QOJJVFM2EUGWH7M7RU.json","graph_json":"https://pith.science/api/pith-number/UPHQIYP2QOJJVFM2EUGWH7M7RU/graph.json","events_json":"https://pith.science/api/pith-number/UPHQIYP2QOJJVFM2EUGWH7M7RU/events.json","paper":"https://pith.science/paper/UPHQIYP2"},"agent_actions":{"view_html":"https://pith.science/pith/UPHQIYP2QOJJVFM2EUGWH7M7RU","download_json":"https://pith.science/pith/UPHQIYP2QOJJVFM2EUGWH7M7RU.json","view_paper":"https://pith.science/paper/UPHQIYP2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2302.01172&json=true","fetch_graph":"https://pith.science/api/pith-number/UPHQIYP2QOJJVFM2EUGWH7M7RU/graph.json","fetch_events":"https://pith.science/api/pith-number/UPHQIYP2QOJJVFM2EUGWH7M7RU/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/UPHQIYP2QOJJVFM2EUGWH7M7RU/action/timestamp_anchor","attest_storage":"https://pith.science/pith/UPHQIYP2QOJJVFM2EUGWH7M7RU/action/storage_attestation","attest_author":"https://pith.science/pith/UPHQIYP2QOJJVFM2EUGWH7M7RU/action/author_attestation","sign_citation":"https://pith.science/pith/UPHQIYP2QOJJVFM2EUGWH7M7RU/action/citation_signature","submit_replication":"https://pith.science/pith/UPHQIYP2QOJJVFM2EUGWH7M7RU/action/replication_record"}},"created_at":"2026-07-05T05:38:23.771326+00:00","updated_at":"2026-07-05T05:38:23.771326+00:00"}