{"record_type":"pith_number_record","schema_url":"https://pith.science/schemas/pith-number/v1.json","pith_number":"pith:2025:VCYOG57QQMMZOIVARMN7WK5QAV","short_pith_number":"pith:VCYOG57Q","schema_version":"1.0","canonical_sha256":"a8b0e377f083199722a08b1bfb2bb0056abbb7c41847000d3a2b8bb5f39c702f","source":{"kind":"arxiv","id":"2502.15470","version":2},"attestation_state":"computed","paper":{"title":"PAPI: Exploiting Dynamic Parallelism in Large Language Model Decoding with a Processing-In-Memory-Enabled Computing System","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.DC","cs.LG"],"primary_cat":"cs.AR","authors_text":"Christina Giannoula, Haiyu Mao, Huawei Li, Juan G\\'omez-Luna, Mohammad Sadrosadati, Onur Mutlu, Xiaowei Li, Ying Wang, Yintao He","submitted_at":"2025-02-21T13:52:31Z","abstract_excerpt":"Large language models (LLMs) are widely used for natural language understanding and text generation. An LLM model relies on a time-consuming step called LLM decoding to generate output tokens. Several prior works focus on improving the performance of LLM decoding using parallelism techniques, such as batching and speculative decoding. State-of-the-art LLM decoding has both compute-bound and memory-bound kernels. Some prior works statically identify and map these different kernels to a heterogeneous architecture consisting of both processing-in-memory (PIM) units and computation-centric acceler"},"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":"2502.15470","kind":"arxiv","version":2},"metadata":{"license":"http://creativecommons.org/licenses/by/4.0/","primary_cat":"cs.AR","submitted_at":"2025-02-21T13:52:31Z","cross_cats_sorted":["cs.AI","cs.DC","cs.LG"],"title_canon_sha256":"485b0fca32489ede5ff34ba2a76da0034358aa2f075c3757ae0b3fc242cf656f","abstract_canon_sha256":"ba03e7b5599434d045bf7950189069574e15517d093873d2d259a136c9b6d7c4"},"schema_version":"1.0"},"receipt":{"kind":"pith_receipt","key_id":"pith-v1-2026-05","algorithm":"ed25519","signed_at":"2026-07-05T10:20:44.724461Z","signature_b64":"iiMkShzCV43pWDZpaNuxQh8pdyccxrnJfQuMaRSZ0s8IAZZXc+ySg6Xl1XYVM6g8STe/2P5ea6hDjG7AN7f1Bw==","signed_message":"canonical_sha256_bytes","builder_version":"pith-number-builder-2026-05-17-v1","receipt_version":"0.3","canonical_sha256":"a8b0e377f083199722a08b1bfb2bb0056abbb7c41847000d3a2b8bb5f39c702f","last_reissued_at":"2026-07-05T10:20:44.723919Z","signature_status":"signed_v1","first_computed_at":"2026-07-05T10:20:44.723919Z","public_key_fingerprint":"8d4b5ee74e4693bcd1df2446408b0d54"},"graph_snapshot":{"paper":{"title":"PAPI: Exploiting Dynamic Parallelism in Large Language Model Decoding with a Processing-In-Memory-Enabled Computing System","license":"http://creativecommons.org/licenses/by/4.0/","headline":"","cross_cats":["cs.AI","cs.DC","cs.LG"],"primary_cat":"cs.AR","authors_text":"Christina Giannoula, Haiyu Mao, Huawei Li, Juan G\\'omez-Luna, Mohammad Sadrosadati, Onur Mutlu, Xiaowei Li, Ying Wang, Yintao He","submitted_at":"2025-02-21T13:52:31Z","abstract_excerpt":"Large language models (LLMs) are widely used for natural language understanding and text generation. An LLM model relies on a time-consuming step called LLM decoding to generate output tokens. Several prior works focus on improving the performance of LLM decoding using parallelism techniques, such as batching and speculative decoding. State-of-the-art LLM decoding has both compute-bound and memory-bound kernels. Some prior works statically identify and map these different kernels to a heterogeneous architecture consisting of both processing-in-memory (PIM) units and computation-centric acceler"},"claims":{"count":0,"items":[],"snapshot_sha256":"258153158e38e3291e3d48162225fcdb2d5a3ed65a07baac614ab91432fd4f57"},"source":{"id":"2502.15470","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/2502.15470/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":"2502.15470","created_at":"2026-07-05T10:20:44.723979+00:00"},{"alias_kind":"arxiv_version","alias_value":"2502.15470v2","created_at":"2026-07-05T10:20:44.723979+00:00"},{"alias_kind":"doi","alias_value":"10.48550/arxiv.2502.15470","created_at":"2026-07-05T10:20:44.723979+00:00"},{"alias_kind":"pith_short_12","alias_value":"VCYOG57QQMMZ","created_at":"2026-07-05T10:20:44.723979+00:00"},{"alias_kind":"pith_short_16","alias_value":"VCYOG57QQMMZOIVA","created_at":"2026-07-05T10:20:44.723979+00:00"},{"alias_kind":"pith_short_8","alias_value":"VCYOG57Q","created_at":"2026-07-05T10:20:44.723979+00:00"}],"events":[],"event_summary":{},"paper_claims":[],"inbound_citations":{"count":1,"internal_anchor_count":0,"sample":[{"citing_arxiv_id":"2505.05772","citing_title":"Sparse Attention Remapping with Clustering for Efficient LLM Decoding on PIM","ref_index":20,"is_internal_anchor":false}]},"formal_canon":{"evidence_count":0,"sample":[],"anchors":[]},"links":{"html":"https://pith.science/pith/VCYOG57QQMMZOIVARMN7WK5QAV","json":"https://pith.science/pith/VCYOG57QQMMZOIVARMN7WK5QAV.json","graph_json":"https://pith.science/api/pith-number/VCYOG57QQMMZOIVARMN7WK5QAV/graph.json","events_json":"https://pith.science/api/pith-number/VCYOG57QQMMZOIVARMN7WK5QAV/events.json","paper":"https://pith.science/paper/VCYOG57Q"},"agent_actions":{"view_html":"https://pith.science/pith/VCYOG57QQMMZOIVARMN7WK5QAV","download_json":"https://pith.science/pith/VCYOG57QQMMZOIVARMN7WK5QAV.json","view_paper":"https://pith.science/paper/VCYOG57Q","resolve_alias":"https://pith.science/api/pith-number/resolve?arxiv=2502.15470&json=true","fetch_graph":"https://pith.science/api/pith-number/VCYOG57QQMMZOIVARMN7WK5QAV/graph.json","fetch_events":"https://pith.science/api/pith-number/VCYOG57QQMMZOIVARMN7WK5QAV/events.json","actions":{"anchor_timestamp":"https://pith.science/pith/VCYOG57QQMMZOIVARMN7WK5QAV/action/timestamp_anchor","attest_storage":"https://pith.science/pith/VCYOG57QQMMZOIVARMN7WK5QAV/action/storage_attestation","attest_author":"https://pith.science/pith/VCYOG57QQMMZOIVARMN7WK5QAV/action/author_attestation","sign_citation":"https://pith.science/pith/VCYOG57QQMMZOIVARMN7WK5QAV/action/citation_signature","submit_replication":"https://pith.science/pith/VCYOG57QQMMZOIVARMN7WK5QAV/action/replication_record"}},"created_at":"2026-07-05T10:20:44.723979+00:00","updated_at":"2026-07-05T10:20:44.723979+00:00"}