{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:NDHSG6IIORM57SA6WMAPCTNM2U","short_pith_number":"pith:NDHSG6II","schema_version":"1.0","canonical_sha256":"68cf2379087459dfc81eb300f14dacd51a6214bda67a30fe9fd5191e8bc82052","source":{"kind":"arxiv","id":"2502.10659","version":1},"attestation_state":"computed","paper":{"title":"Pushing up to the Limit of Memory Bandwidth and Capacity Utilization for Efficient LLM Decoding on Embedded FPGA","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AR","authors_text":"Dongcheng Zhao, Guobin Shen, Jindong Li, Qian Zhang, Tenglong Li, Yi Zeng","submitted_at":"2025-02-15T03:56:22Z","abstract_excerpt":"The extremely high computational and storage demands of large language models have excluded most edge devices, which were widely used for efficient machine learning, from being viable options. A typical edge device usually only has 4GB of memory capacity and a bandwidth of less than 20GB/s, while a large language model quantized to 4-bit precision with 7B parameters already requires 3.5GB of capacity, and its decoding process is purely bandwidth-bound. In this paper, we aim to explore these limits by proposing a hardware accelerator for large language model (LLM) inference on the Zynq-based KV"},"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":"2502.10659","kind":"arxiv","version":1},"metadata":{"license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","primary_cat":"cs.AR","submitted_at":"2025-02-15T03:56:22Z","cross_cats_sorted":[],"title_canon_sha256":"0851399bfbc3a4afa3c5a2ac6080ad9c0c6d690ea74bc0444f8251cd52d950d8","abstract_canon_sha256":"e1e71146ce597e3d6224bf52fb8b8c08a9ff029d7de97c8188ca1761ed6160dd"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:14:55.897743Z","signature_b64":"O+ID00nijxlUUbA2uQ5YElD3ABy8DUtiNPAhIG/+HNzpgNPtGVqKQ4j5/fUF+uAOKJ1+LsZI3N1pwnKxShkBAw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"68cf2379087459dfc81eb300f14dacd51a6214bda67a30fe9fd5191e8bc82052","last_reissued_at":"2026-07-05T10:14:55.897239Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:14:55.897239Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"Pushing up to the Limit of Memory Bandwidth and Capacity Utilization for Efficient LLM Decoding on Embedded FPGA","license":"http://arxiv.org/licenses/nonexclusive-distrib/1.0/","headline":"","cross_cats":[],"primary_cat":"cs.AR","authors_text":"Dongcheng Zhao, Guobin Shen, Jindong Li, Qian Zhang, Tenglong Li, Yi Zeng","submitted_at":"2025-02-15T03:56:22Z","abstract_excerpt":"The extremely high computational and storage demands of large language models have excluded most edge devices, which were widely used for efficient machine learning, from being viable options. A typical edge device usually only has 4GB of memory capacity and a bandwidth of less than 20GB/s, while a large language model quantized to 4-bit precision with 7B parameters already requires 3.5GB of capacity, and its decoding process is purely bandwidth-bound. In this paper, we aim to explore these limits by proposing a hardware accelerator for large language model (LLM) inference on the Zynq-based KV"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.10659","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/2502.10659/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":"2502.10659","created_at":"2026-07-05T10:14:55.897299+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.10659v1","created_at":"2026-07-05T10:14:55.897299+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.10659","created_at":"2026-07-05T10:14:55.897299+00:00"},{"alias_kind":"pith_short_12","alias_value":"NDHSG6IIORM5","created_at":"2026-07-05T10:14:55.897299+00:00"},{"alias_kind":"pith_short_16","alias_value":"NDHSG6IIORM57SA6","created_at":"2026-07-05T10:14:55.897299+00:00"},{"alias_kind":"pith_short_8","alias_value":"NDHSG6II","created_at":"2026-07-05T10:14:55.897299+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":1,"sample":[{"citing_arxiv_id":"2505.16335","citing_title":"FPQVAR: Floating Point Quantization for Visual Autoregressive Model with FPGA Hardware Co-design","ref_index":24,"is_internal_anchor":true}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/NDHSG6IIORM57SA6WMAPCTNM2U","json":"https://pith.science/pith/NDHSG6IIORM57SA6WMAPCTNM2U.json","graph_json":"https://pith.science/api/pith-number/NDHSG6IIORM57SA6WMAPCTNM2U/graph.json","events_json":"https://pith.science/api/pith-number/NDHSG6IIORM57SA6WMAPCTNM2U/events.json","paper":"https://pith.science/paper/NDHSG6II"},"agent_actions":{"view_html":"https://pith.science/pith/NDHSG6IIORM57SA6WMAPCTNM2U","download_json":"https://pith.science/pith/NDHSG6IIORM57SA6WMAPCTNM2U.json","view_paper":"https://pith.science/paper/NDHSG6II","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.10659&json=true","fetch_graph":"https://pith.science/api/pith-number/NDHSG6IIORM57SA6WMAPCTNM2U/graph.json","fetch_events":"https://pith.science/api/pith-number/NDHSG6IIORM57SA6WMAPCTNM2U/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/NDHSG6IIORM57SA6WMAPCTNM2U/action/timestamp_anchor","attest_storage":"https://pith.science/pith/NDHSG6IIORM57SA6WMAPCTNM2U/action/storage_attestation","attest_author":"https://pith.science/pith/NDHSG6IIORM57SA6WMAPCTNM2U/action/author_attestation","sign_citation":"https://pith.science/pith/NDHSG6IIORM57SA6WMAPCTNM2U/action/citation_signature","submit_replication":"https://pith.science/pith/NDHSG6IIORM57SA6WMAPCTNM2U/action/replication_record"}},"created_at":"2026-07-05T10:14:55.897299+00:00","updated_at":"2026-07-05T10:14:55.897299+00:00"}