{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2026:BL5OZJSCI5YHCIJAGIO7CCMGWC","short_pith_number":"pith:BL5OZJSC","schema_version":"1.0","canonical_sha256":"0afaeca6424770712120321df10986b09913f4d0e7ac6937e3616e8b3d594dd2","source":{"kind":"arxiv","id":"2602.04595","version":3},"attestation_state":"computed","paper":{"title":"Harmonia: Algorithm-Hardware Co-Design for Memory- and Compute-Efficient BFP-based LLM Inference","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AR","authors_text":"Jieyu Li, Weifeng He, Xinyu Wang, Yanan Sun","submitted_at":"2026-02-04T14:22:08Z","abstract_excerpt":"Large Language Models (LLMs) incur substantial memory and computation costs. Prior works reduce FP-INT arithmetic overhead by converting linear-layer activations to block floating point (BFP), but retain FP activations in attention layers due to accuracy concerns. We propose Harmonia, an algorithm-hardware co-design framework that enables BFP representation and computation across both linear and attention layers. Harmonia first explores BFP configurations to balance model accuracy and activation compression. It then combines asymmetric bit allocation with hybrid offline-online outlier smoothin"},"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":"2602.04595","kind":"arxiv","version":3},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AR","submitted_at":"2026-02-04T14:22:08Z","cross_cats_sorted":[],"title_canon_sha256":"8322bb1c3b1c347d1aef6250b9e9e70fc4f874c45992ffbf505f96120b8c3d32","abstract_canon_sha256":"289c06899dd41317bfa8139d9beb6525f239b8280cf3c8166405fc9d69b23ae2"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-23T01:24:21.474738Z","signature_b64":"ttcQK9kBCl6zwMOrGdp+ob1QA6pFqGoPZd4Mf2NheVFuY/E5jqiNNdMjM1yQ0hSmk/BGoLt7bAWlB0QbJh/ADQ==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"0afaeca6424770712120321df10986b09913f4d0e7ac6937e3616e8b3d594dd2","last_reissued_at":"2026-07-23T01:24:21.473771Z","signature_status":"signed_v1","first_computed_at":"2026-07-23T01:24:21.473771Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Harmonia: Algorithm-Hardware Co-Design for Memory- and Compute-Efficient BFP-based LLM Inference","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AR","authors_text":"Jieyu Li, Weifeng He, Xinyu Wang, Yanan Sun","submitted_at":"2026-02-04T14:22:08Z","abstract_excerpt":"Large Language Models (LLMs) incur substantial memory and computation costs. Prior works reduce FP-INT arithmetic overhead by converting linear-layer activations to block floating point (BFP), but retain FP activations in attention layers due to accuracy concerns. We propose Harmonia, an algorithm-hardware co-design framework that enables BFP representation and computation across both linear and attention layers. Harmonia first explores BFP configurations to balance model accuracy and activation compression. It then combines asymmetric bit allocation with hybrid offline-online outlier smoothin"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2602.04595","kind":"arxiv","version":3},"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/2602.04595/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":"2602.04595","created_at":"2026-07-23T01:24:21.474207+00:00"},{"alias_kind":"arxiv_version","alias_value":"2602.04595v3","created_at":"2026-07-23T01:24:21.474207+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2602.04595","created_at":"2026-07-23T01:24:21.474207+00:00"},{"alias_kind":"pith_short_12","alias_value":"BL5OZJSCI5YH","created_at":"2026-07-23T01:24:21.474207+00:00"},{"alias_kind":"pith_short_16","alias_value":"BL5OZJSCI5YHCIJA","created_at":"2026-07-23T01:24:21.474207+00:00"},{"alias_kind":"pith_short_8","alias_value":"BL5OZJSC","created_at":"2026-07-23T01:24:21.474207+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":0,"internal_anchor_count":0,"sample":[]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/BL5OZJSCI5YHCIJAGIO7CCMGWC","json":"https://pith.science/pith/BL5OZJSCI5YHCIJAGIO7CCMGWC.json","graph_json":"https://pith.science/api/pith-number/BL5OZJSCI5YHCIJAGIO7CCMGWC/graph.json","events_json":"https://pith.science/api/pith-number/BL5OZJSCI5YHCIJAGIO7CCMGWC/events.json","paper":"https://pith.science/paper/BL5OZJSC"},"agent_actions":{"view_html":"https://pith.science/pith/BL5OZJSCI5YHCIJAGIO7CCMGWC","download_json":"https://pith.science/pith/BL5OZJSCI5YHCIJAGIO7CCMGWC.json","view_paper":"https://pith.science/paper/BL5OZJSC","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2602.04595&json=true","fetch_graph":"https://pith.science/api/pith-number/BL5OZJSCI5YHCIJAGIO7CCMGWC/graph.json","fetch_events":"https://pith.science/api/pith-number/BL5OZJSCI5YHCIJAGIO7CCMGWC/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/BL5OZJSCI5YHCIJAGIO7CCMGWC/action/timestamp_anchor","attest_storage":"https://pith.science/pith/BL5OZJSCI5YHCIJAGIO7CCMGWC/action/storage_attestation","attest_author":"https://pith.science/pith/BL5OZJSCI5YHCIJAGIO7CCMGWC/action/author_attestation","sign_citation":"https://pith.science/pith/BL5OZJSCI5YHCIJAGIO7CCMGWC/action/citation_signature","submit_replication":"https://pith.science/pith/BL5OZJSCI5YHCIJAGIO7CCMGWC/action/replication_record"}},"created_at":"2026-07-23T01:24:21.474207+00:00","updated_at":"2026-07-23T01:24:21.474207+00:00"}