{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:APHFF3K2MSAXFMSTPDKRFERIFZ","short_pith_number":"pith:APHFF3K2","schema_version":"1.0","canonical_sha256":"03ce52ed5a648172b25378d51292282e7b7000c1e288020d3c46a1853d6f249d","source":{"kind":"arxiv","id":"2504.09064","version":1},"attestation_state":"computed","paper":{"title":"PQS (Prune, Quantize, and Sort): Low-Bitwidth Accumulation of Dot Products in Neural Network Computations","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"H.T. Kung, Vikas Natesh","submitted_at":"2025-04-12T03:51:42Z","abstract_excerpt":"We present PQS, which uses three techniques together - Prune, Quantize, and Sort - to achieve low-bitwidth accumulation of dot products in neural network computations. In conventional quantized (e.g., 8-bit) dot products, partial results are accumulated into wide (e.g., 32-bit) accumulators to avoid overflows when accumulating intermediate partial sums. However, such wide accumulators increase memory bandwidth usage and reduce energy efficiency. We show that iterative N:M pruning in floating point followed by quantization to 8 (or fewer) bits, and accumulation of partial products in a sorted o"},"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":"2504.09064","kind":"arxiv","version":1},"metadata":{"license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","primary_cat":"cs.LG","submitted_at":"2025-04-12T03:51:42Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"ee038f025be071426a5b7e6127042775bab9895d03030290a8e48e4652e6e3d0","abstract_canon_sha256":"e9e59362ea6529ffe3d264d4fbade3e8452034f4f0aafd805a2680e5cc6c8513"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:48:03.604028Z","signature_b64":"066iJngUCpn2XnTAzCUWs417GNhSAXN6YyEQIdU4ej4v5oY4xQdm3YIYURwWL75/ScbiOmxh4SiDVaJ9flJnBg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"03ce52ed5a648172b25378d51292282e7b7000c1e288020d3c46a1853d6f249d","last_reissued_at":"2026-07-05T10:48:03.603558Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:48:03.603558Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"PQS (Prune, Quantize, and Sort): Low-Bitwidth Accumulation of Dot Products in Neural Network Computations","license":"http://creativecommons.org/licenses/by-nc-nd/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.LG","authors_text":"H.T. Kung, Vikas Natesh","submitted_at":"2025-04-12T03:51:42Z","abstract_excerpt":"We present PQS, which uses three techniques together - Prune, Quantize, and Sort - to achieve low-bitwidth accumulation of dot products in neural network computations. In conventional quantized (e.g., 8-bit) dot products, partial results are accumulated into wide (e.g., 32-bit) accumulators to avoid overflows when accumulating intermediate partial sums. However, such wide accumulators increase memory bandwidth usage and reduce energy efficiency. We show that iterative N:M pruning in floating point followed by quantization to 8 (or fewer) bits, and accumulation of partial products in a sorted o"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2504.09064","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/2504.09064/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":"2504.09064","created_at":"2026-07-05T10:48:03.603620+00:00"},{"alias_kind":"arxiv_version","alias_value":"2504.09064v1","created_at":"2026-07-05T10:48:03.603620+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2504.09064","created_at":"2026-07-05T10:48:03.603620+00:00"},{"alias_kind":"pith_short_12","alias_value":"APHFF3K2MSAX","created_at":"2026-07-05T10:48:03.603620+00:00"},{"alias_kind":"pith_short_16","alias_value":"APHFF3K2MSAXFMST","created_at":"2026-07-05T10:48:03.603620+00:00"},{"alias_kind":"pith_short_8","alias_value":"APHFF3K2","created_at":"2026-07-05T10:48:03.603620+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2506.11728","citing_title":"The Cambrian Explosion of Mixed-Precision Matrix Multiplication for Quantized Deep Learning Inference","ref_index":24,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/APHFF3K2MSAXFMSTPDKRFERIFZ","json":"https://pith.science/pith/APHFF3K2MSAXFMSTPDKRFERIFZ.json","graph_json":"https://pith.science/api/pith-number/APHFF3K2MSAXFMSTPDKRFERIFZ/graph.json","events_json":"https://pith.science/api/pith-number/APHFF3K2MSAXFMSTPDKRFERIFZ/events.json","paper":"https://pith.science/paper/APHFF3K2"},"agent_actions":{"view_html":"https://pith.science/pith/APHFF3K2MSAXFMSTPDKRFERIFZ","download_json":"https://pith.science/pith/APHFF3K2MSAXFMSTPDKRFERIFZ.json","view_paper":"https://pith.science/paper/APHFF3K2","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2504.09064&json=true","fetch_graph":"https://pith.science/api/pith-number/APHFF3K2MSAXFMSTPDKRFERIFZ/graph.json","fetch_events":"https://pith.science/api/pith-number/APHFF3K2MSAXFMSTPDKRFERIFZ/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/APHFF3K2MSAXFMSTPDKRFERIFZ/action/timestamp_anchor","attest_storage":"https://pith.science/pith/APHFF3K2MSAXFMSTPDKRFERIFZ/action/storage_attestation","attest_author":"https://pith.science/pith/APHFF3K2MSAXFMSTPDKRFERIFZ/action/author_attestation","sign_citation":"https://pith.science/pith/APHFF3K2MSAXFMSTPDKRFERIFZ/action/citation_signature","submit_replication":"https://pith.science/pith/APHFF3K2MSAXFMSTPDKRFERIFZ/action/replication_record"}},"created_at":"2026-07-05T10:48:03.603620+00:00","updated_at":"2026-07-05T10:48:03.603620+00:00"}