{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2024:JNDKID67RZ2DDFMISHSELLTN7U","short_pith_number":"pith:JNDKID67","schema_version":"1.0","canonical_sha256":"4b46a40fdf8e7431958891e445ae6dfd2232e4f1b18a4f4001c84d68818186b9","source":{"kind":"arxiv","id":"2402.00025","version":2},"attestation_state":"computed","paper":{"title":"Accelerating a Triton Fused Kernel for W4A16 Quantized Inference with SplitK work decomposition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.DC","authors_text":"Adnan Hoque, Chih-Chieh Yang, Less Wright, Mudhakar Srivatsa, Raghu Ganti","submitted_at":"2024-01-05T19:17:55Z","abstract_excerpt":"We propose an implementation of an efficient fused matrix multiplication kernel for W4A16 quantized inference, where we perform dequantization and GEMM in a fused kernel using a SplitK work decomposition. Our implementation shows improvement for the type of skinny matrix-matrix multiplications found in foundation model inference workloads. In particular, this paper surveys the type of matrix multiplication between a skinny activation matrix and a square weight matrix. Our results show an average of 65% speed improvement on A100, and an average of 124% speed improvement on H100 (with a peak 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":"2402.00025","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.DC","submitted_at":"2024-01-05T19:17:55Z","cross_cats_sorted":["cs.AI"],"title_canon_sha256":"2f730c10376a3cbab5771346e0daf7dffb516d705214dd42ff89d23ffeb2b7db","abstract_canon_sha256":"5a2b3c2621efdfee38b0d3fd93d1257f23448152c17c09a0640c0f16da189999"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T07:48:17.924701Z","signature_b64":"Irkd4Ayz74j0DEpZVYrndRocQAv2w7ZoYfYHP4QFyIzQJvyglRuPAljQ885MjzyjZrYfTQ/F3rdoOQSs2UgoDg==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"4b46a40fdf8e7431958891e445ae6dfd2232e4f1b18a4f4001c84d68818186b9","last_reissued_at":"2026-07-05T07:48:17.924202Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T07:48:17.924202Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Accelerating a Triton Fused Kernel for W4A16 Quantized Inference with SplitK work decomposition","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI"],"primary_cat":"cs.DC","authors_text":"Adnan Hoque, Chih-Chieh Yang, Less Wright, Mudhakar Srivatsa, Raghu Ganti","submitted_at":"2024-01-05T19:17:55Z","abstract_excerpt":"We propose an implementation of an efficient fused matrix multiplication kernel for W4A16 quantized inference, where we perform dequantization and GEMM in a fused kernel using a SplitK work decomposition. Our implementation shows improvement for the type of skinny matrix-matrix multiplications found in foundation model inference workloads. In particular, this paper surveys the type of matrix multiplication between a skinny activation matrix and a square weight matrix. Our results show an average of 65% speed improvement on A100, and an average of 124% speed improvement on H100 (with a peak of "},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2402.00025","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/2402.00025/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":"2402.00025","created_at":"2026-07-05T07:48:17.924260+00:00"},{"alias_kind":"arxiv_version","alias_value":"2402.00025v2","created_at":"2026-07-05T07:48:17.924260+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2402.00025","created_at":"2026-07-05T07:48:17.924260+00:00"},{"alias_kind":"pith_short_12","alias_value":"JNDKID67RZ2D","created_at":"2026-07-05T07:48:17.924260+00:00"},{"alias_kind":"pith_short_16","alias_value":"JNDKID67RZ2DDFMI","created_at":"2026-07-05T07:48:17.924260+00:00"},{"alias_kind":"pith_short_8","alias_value":"JNDKID67","created_at":"2026-07-05T07:48:17.924260+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2606.01117","citing_title":"HASTE: Hardware-Aware Dynamic Sparse Training for Large Output Spaces","ref_index":38,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/JNDKID67RZ2DDFMISHSELLTN7U","json":"https://pith.science/pith/JNDKID67RZ2DDFMISHSELLTN7U.json","graph_json":"https://pith.science/api/pith-number/JNDKID67RZ2DDFMISHSELLTN7U/graph.json","events_json":"https://pith.science/api/pith-number/JNDKID67RZ2DDFMISHSELLTN7U/events.json","paper":"https://pith.science/paper/JNDKID67"},"agent_actions":{"view_html":"https://pith.science/pith/JNDKID67RZ2DDFMISHSELLTN7U","download_json":"https://pith.science/pith/JNDKID67RZ2DDFMISHSELLTN7U.json","view_paper":"https://pith.science/paper/JNDKID67","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2402.00025&json=true","fetch_graph":"https://pith.science/api/pith-number/JNDKID67RZ2DDFMISHSELLTN7U/graph.json","fetch_events":"https://pith.science/api/pith-number/JNDKID67RZ2DDFMISHSELLTN7U/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/JNDKID67RZ2DDFMISHSELLTN7U/action/timestamp_anchor","attest_storage":"https://pith.science/pith/JNDKID67RZ2DDFMISHSELLTN7U/action/storage_attestation","attest_author":"https://pith.science/pith/JNDKID67RZ2DDFMISHSELLTN7U/action/author_attestation","sign_citation":"https://pith.science/pith/JNDKID67RZ2DDFMISHSELLTN7U/action/citation_signature","submit_replication":"https://pith.science/pith/JNDKID67RZ2DDFMISHSELLTN7U/action/replication_record"}},"created_at":"2026-07-05T07:48:17.924260+00:00","updated_at":"2026-07-05T07:48:17.924260+00:00"}